Long-term in vivo three-photon imaging reveals region-specific differences in healthy and regenerative oligodendrogenesis

Authors: Michael A. Thornton, Gregory L. Futia, Michael E. Stockton, Samuel A. Budoff, Alexandra N Ramirez, Baris Ozbay, Omer Tzang, Karl Kilborn, Alon Poleg-Polsky, Diego Restrepo, Emily A. Gibson, Ethan G. Hughes

Abstract

The generation of new myelin-forming oligodendrocytes in the adult CNS is critical for cognitive function and regeneration following injury. Oligodendrogenesis varies between gray and white matter regions suggesting that local cues drive regional differences in myelination and the capacity for regeneration. Yet, the determination of regional variability in oligodendrocyte cell behavior is limited by the inability to monitor the dynamics of oligodendrocytes and their transcriptional subpopulations in white matter of the living brain. Here, we harnessed the superior imaging depth of three-photon microscopy to permit long-term, longitudinal in vivo three-photon imaging of an entire cortical column and underlying subcortical white matter without cellular damage or reactivity. Using this approach, we found that the white matter generated substantially more new oligodendrocytes per volume compared to the gray matter, yet the rate of population growth was proportionally higher in the gray matter. Following demyelination, the white matter had an enhanced population growth that resulted in higher oligodendrocyte replacement compared to the gray matter. Finally, deep cortical layers had pronounced deficits in regenerative oligodendrogenesis and restoration of the MOL5/6-positive oligodendrocyte subpopulation following demyelinating injury. Together, our findings demonstrate that regional microenvironments regulate oligodendrocyte population dynamics and heterogeneity in the healthy and diseased brain.

In the central nervous system (CNS), oligodendrocytes produce myelin, which enwraps neuronal axons to increase axonal conduction, fine-tune circuit timing, and provide metabolic support. In 1901, Paul Flechsig proposed that regional myelination is regulated by the functional maturation of neural circuits (fundamental law of myelogenesis^1^), implying that the cellular microenvironment may shape the generation of oligodendrocytes.

Throughout life, oligodendrocytes are generated via the differentiation of oligodendrocyte precursor cells (OPCs, also called oligodendrocyte progenitor cells). Recent work shows that environmental factors such as brain region and age influence OPCs to become functionally heterogenous via differential ion channel expression^2^. In line with these findings, the rates of OPC proliferation and differentiation differ between gray and white matter areas^3,4^ and white matter-derived precursors generate oligodendrocytes with a higher efficiency than gray matter-derived OPCs after transplantation^5^. Additionally, mature oligodendrocytes make up a heterogeneous cell population with distinct transcriptional subtypes^6^ that exhibit spatial preference^7,8^ and differential responses to injury and disease^9-11^. However, a complete understanding of region-specific regulation of oligodendrogenesis and transcriptional heterogeneity is limited due to the inability to monitor the population dynamics of oligodendrocytes in white matter in the living brain.

In demyelinating diseases, such as multiple sclerosis (MS), death of oligodendrocytes elicits regenerative oligodendrogenesis^12^ and remyelination that can restore neuronal function^13,14^. Historically, MS has been regarded as a disease of the white matter; however, recent work shows that MS patients have extensive gray matter lesions that correlate with late-stage disabilities^15,16^. Interestingly, leukocortical lesions that span white and gray matter regions display distinct pathological hallmarks suggesting that cellular microenvironments of lesions in different brain regions distinctly regulate myelin loss and repair^17^. Yet, the cellular dynamics of oligodendrocyte population loss and regeneration following demyelination remain uncharacterized across brain regions.

Advancements in multiphoton microscopy have revolutionized our understanding of cellular dynamics in intact tissues by extending imaging depths four-fold compared to conventional confocal microscopy^18^. For example, previous studies using longitudinal in vivo two-photon imaging in mouse primary motor cortex^14^ and somatosensory cortex^12,19,20^ revealed fundamental aspects of the dynamics of oligodendrocytes in the healthy and demyelinated brain. However, due to light scattering in brain tissue these studies have been limited to the superficial ~400 microns of the cortical gray matter when extended over multiple months. The recent development and application of three-photon microscopy, which uses a longer excitation wavelength and experiences reduced scattering compared to two-photon, has permitted in vivo imaging depths greater than 1 mm at high resolution^21-23^. However, the high peak intensity of excitation pulses required for nonlinear excitation can result in tissue damage particularly when applied repetitively^24^. Whether three-photon microscopy can be reliably utilized for long-term volumetric measurements of cellular behavior in the intact brain remains unclear.

Here, we asked whether the dynamics of oligodendrocyte populations differ across cortical and subcortical regions in the context of health and disease. To address this question, we developed an imaging protocol that permits long-term volumetric in vivo three-photon imaging without tissue damage, cellular reactivity, or perturbation of healthy oligodendrogenesis in the adult mouse brain. This protocol enabled us to track the cellular behavior of over 13,000 individual oligodendrocytes across 2-3 months in health and disease. Long-term in vivo three-photon imaging showed that white matter generated more oligodendrocytes per volume compared to gray matter, however, the population growth of oligodendrocytes was accelerated in the gray matter. Following cuprizone-mediated demyelination, we found that the oligodendrocytes and their subpopulations were equivalently reduced across gray and white matter, however, the white matter showed enhanced regenerative oligodendrogenesis and population replacement compared to gray matter regions. Layer-specific analyses revealed that the deep cortical layers had a reduced capacity for regenerative oligodendrogenesis and restoration of the MOL5/6 oligodendrocyte subpopulation following demyelinating injury. Together, these results indicate that regional microenvironments differentially modulate oligodendrogenesis and heterogeneous oligodendrocyte subpopulations to influence myelination and regeneration in the healthy and diseased brain.

Results

In vivo three-photon imaging of cortical gray and white matter

To determine region-specific dynamics of oligodendrocyte generation in an entire cortical column encompassing the six layers of the cerebral cortex and the subcortical white matter, we built a custom three-photon microscope that enabled imaging to depths of ~1000 μm across multiple months (Methods, Extended Data Fig. 1). We implanted chronic cranial windows over the posterior parietal cortex (PPC) of MOBP-EGFP transgenic mice^25^ where enhanced green fluorescent protein (EGFP) expression is driven by the MOBP promoter, thereby labeling all mature oligodendrocytes and their associated myelin sheaths in both the gray and white matter (Extended Data Fig. 2). To determine the efficacy of three-photon microscopy to visualize oligodendrocytes across cortical gray and white matter regions, we compared two-photon and three-photon imaging modalities. Two-photon imaging of the highest clarity cranial windows in MOBP-EGFP transgenic mice three weeks after implantation allowed the visualization of oligodendrocyte cell bodies to ~600 μm from the brain surface (Fig. 1a). Three-photon imaging allowed for the collection of ~400 x 400 x 1050 μm 3D volumes of EGFP-labeled oligodendrocytes encompassing the entire cortical column as well as subcortical white matter in chronic cranial windows of a wide range of clarity (Fig. 1b, Supplementary Video 1). Because multiphoton imaging quality is dependent on a combination of signal generation and out-of-focus background, we used the signal-to-background ratio (SBR) to compare two-photon and three-photon image quality in MOBP-EGFP mice. We found that two-photon imaging had higher SBR of oligodendrocyte cell bodies compared to three-photon imaging in the superficial 100 μm of cortex. This enhanced SBR is likely due to the increased probability of two-photon excitation in regions of minimal scattering^24^. In contrast, we found that three-photon imaging had increased SBR at depths greater than 400 μm beneath the brain surface compared to two-photon imaging (21.8±3.9 vs. 7.3±3.1 for 401 – 500 μm depth, Fig. 1c). This increase in three-photon image quality resulted in the detection of a higher number of oligodendrocytes in layer 5 when compared with two-photon data (78.8±10.2 vs. 41.4±10.7, Fig. 1d), highlighting that two-photon imaging only partially samples the labeled oligodendrocyte population at depths below superficial cortical layers 1–4.

Three-photon microscopy allowed for the collection of third harmonic generation (THG) signal, which is generated at optical found that repeated, long-term three-photon volumetric imaging using these parameters caused cumulative tissue damage that was visualized using THG imaging (Fig. 2a). This laser-induced damage resulted in oligodendrocyte cell death and a subsequent burst of oligodendrogenesis that could be predicted by a two-fold increase in the THG intensity at the previous time point (Fig. 2b-c). The onset of diffuse 1300 nm excitation-induced oligodendrocyte cell death was more rapid than reported for targeted single-oligodendrocyte ablation with 775 nm excitation^32^, and was sustained for up to fifty days. The injury-induced oligodendrogenesis had variable dynamics with a peak rate at ~20-40 days post-lesion (Fig. 2b,c). To develop an imaging protocol appropriate for multi-month longitudinal in vivo imaging of cell behavior, we implemented several optical and scanning modifications (Extended Data Fig. 4). First, to reduce optical aberrations and increase image quality, we used dual-axis goniometers to align the chronic cranial window to be perpendicular to the light path^33^. Second, the excitation laser was blanked on the galvonometer over-scan to reduce the average power applied to the brain to < 52 mW (Extended Data Fig. 4). Third, frame averaging was utilized rather than pixel averaging or extended dwell time techniques, and the acquisition was paused for one minute per every three minutes of continuous scanning Fourth, we used adaptive optics optimized on the THG signal just above the corpus callosum to enhance the SBR of deep interfaces of different refractive index and third-order nonlinear susceptibility^26,27^ (Supplementary Video 1). Similar to previous studies^21,28^, we found that THG strongly labeled blood vessels from venules to capillaries, individual myelin sheaths, and myelinated fiber bundles in the subcortical white matter (Fig. 1b). We confirmed that THG-positive sheaths in the superficial PPC colocalized with EGFP+ sheaths in MOBP-EGFP transgenic mice and revealed a similar in vivo myelin labeling pattern to Spectral Confocal Reflectance (SCoRe) microscopy^29^ (Extended Data Fig. 2). Adaptive optics (AO) via image-based Zernike-mode aberration correction with a deformable mirror revealed individual THG-positive myelin sheath cross sections at depths of up to 1000 μm and increased the SBR and axial resolution of oligodendrocytes in the corpus callosum (Extended Data Fig. 3). At depths of 750 – 900 μm, AO correction resulted in a 68% increase in the normalized 3P signal of oligodendrocyte cell bodies and a 202% increase in oligodendrocyte processes. Furthermore, AO optimization allowed for a ~56% reduction in the average power applied to the sample at depth (Extended Data Fig. 3). Additionally, we used THG microscopy to determine the depth from cortical surface to white matter and to further define anatomical landmarks within the white matter. Across eight, ten-week-old mice, we found that the average distance from brain surface to white matter in the PPC was 823±16 μm, while the thickness of the white matter was 118±8 μm, for a total imaging depth of 941±11 μm (Fig. 1e). The subcortical white matter below the posterior parietal cortex consists of myelinated fiber tracts of output axons from the cortex (corpus callosum) and the hippocampus (alveus)^30^ (Fig. 1f). We found that the orientation of the THG+ fibers allowed for the delineation of the corpus callosum and the alveus of the hippocampal formation (15.3±4.9 vs. −0.4±1.04 degrees, respectively, Fig. 1b,f). Together, these data show that three-photon imaging provides greater imaging depth, higher imaging quality, and an additional imaging modality to study oligodendrocytes and myelination in cortical gray and white matter regions.

Optimization of excitation parameters for long-term three-photon microscopy without tissue damage

Although previously published three-photon average power guidelines^31^ allowed for successful acquisition of single time-point three-dimensional volumetric and continuous single-plane time lapse imaging (Fig. 1), we oligodendrocytes (Extended Data Fig. 4), and linearly modulated the spherical correction with depth to maintain SBR in more superficial regions. Due to mouse-to-mouse variability in the scattering properties of the intact brain, nearly identical exponential power curves resulted in either healthy long-term imaging or lesions with oligodendrocyte cell death (Fig. 2d). Therefore, we measured mouse-specific effective attenuation lengths (EALs) in the adult posterior parietal cortex to estimate the pulse energy delivered at each focal plane as described previously^21,28,34^ (Extended Data Figure 5; Methods). The pulse energy at the focal plane was calculated using the equation

Pf=Pse−(zEAL)

Where Pf is pulse energy at the focal plane (nJ), Ps is pulse energy at the surface (nJ), Z is depth from the brain surface (μm), EAL is the measured effective attenuation lengths (gray and white matter) for each mouse. Maintaining a relatively constant pulse energy of ~2 nJ across the cortical depth (1.98±0.20 nJ; n = 11 longitudinally-imaged mice) with an absolute maximum threshold of 2.7 nJ (Fig. 2e) was sufficient to maintain high oligodendrocyte SBR and track healthy oligodendrogenesis without oligodendrocyte death (Fig. 2f). The average power modulation curves varied slightly for each mouse due to differences in the thickness and opacity of the meninges at the surface of the brain under chronic cranial windows (Extended Data Fig. 5). The average power modulation by depth curves that were applied over multiple months of 3P imaging without damage can be described by the exponential

APs∼1.5e0.0036z

Where is APs is Average Power at the Surface (mW) and z is Z-depth from the brain surface (μm) (Extended Data Fig. 5). The entire depth of the cortex, excluding the white matter, could be imaged with less than 30 mW of average power at the surface.

Long-term in vivo three-photon imaging does not cause tissue damage or cellular reactivity

To determine whether these modifications permitted long-term in vivo three-photon imaging without tissue damage, we assessed cellular reactivity, phototoxicity, and molecular stress via immunohistochemistry in tissue from mice that were exposed to three-photon longitudinal imagining for over 2.5 months. Mice were perfused immediately following the final day of in vivo imaging and the 385 x 385 x 1000 μm imaged regions were localized, processed for immunofluorescence, and compared to the contralateral cerebral hemisphere as described previously^31,35^ (Fig. 3, Extended Data Fig. 6). To assess the sensitivity and specificity of our histological damage readouts, in a separate group of mice, we induced laser lesions by saturating the THG channel at each z-plane during 3D acquisition and perfused the mice 18-24 hours later (Fig. 3a,b, Extended Data Fig. 6). These laser-induced lesions displayed an increase in the diffuse extracellular THG signal and tissue opacity, suggesting that we induced mild nonlinear- and heating-related damage (Fig. 2). First, we found no differences in the cellular reactivity of oligodendrocytes, microglia, or astrocytes between the imaged and contralateral hemispheres in healthy mice that received 2.5 months of optimized long-term 3P imaging (Extended Data Fig. 6). In contrast, positive control tissue, showed a marked increase in the density of microglia and astrocytes in the lesion sites (440.7±45.6 vs. 262.2±18.3 microglia / mm^2^; 290.8±76.4 vs. 142.4±49.7 astrocytes / mm^2^, Extended Data Fig. 6). Analyses of the RNA oxidation marker 8-hydroxyguanosine (8-OHG)^36^ did not reveal significant oxidative stress responses in neurons to either type of laser exposure (Extended Data Fig. 6). Next, we analyzed blood vessel coverage and pericyte density (Lectin-649, vasculature; CD13, pericytes) using automated segmentation^37,38^ and found no effects of long-term three-photon imaging (Extended Data Fig. 6). Interestingly, blood vessel coverage increased in acutely damaged positive control mice (22.8±3.4 vs. 16.8±2.3%) suggesting a dilation of the vasculature in response to this type of photodamage. Pericyte density was unchanged following laser injury mice, confirming that this type of photodamage does not cause cellular ablation. Together, these results suggest that while long-term, repeated three-photon excitation has the potential to induce phototoxicity, our engineering and operational controls allow for multi-month longitudinal imaging without the induction of cellular reactivity at the tissue level.

Long-term three-photon imaging does not affect healthy oligodendrogenesis or cell death

Next, we examined if long-term three-photon imaging induced molecular stress or cellular behavioral changes specifically in mature oligodendrocytes. To examine oligodendrocyte molecular stress, we segmented EGFP-positive cell bodies in MOBP-EGFP transgenic tissue from long-term three-photon and acute laser injury mice and counted the density of oligodendrocytes that were positive for HSP70/72, a known marker of thermal stress, and γ-H2A.X, a marker induced by DNA damage (Fig. 3c,d; Methods). We found no differences in HSP70/72- or γ-H2A.X-positive oligodendrocytes between the imaged and contralateral cerebral hemispheres following long-term three-photon imaging (Fig. 3e,h). Acute laser injury tissue showed an increase in the density of HSP70/72-positive oligodendrocytes and the fluorescence intensity on the exposed hemisphere (71.7±22.6 vs. 39.2±18.6 HSP+ OLs/mm^2^; 0.91±0.08 (healthy) vs. 5.26±2.46 (laser-injury) Fig. 3f,g), confirming that oligodendrocytes respond rapidly to thermal stress, yet the density and intensity of γ-H2A.X-positive oligodendrocytes did not reach statistical significance (Fig. 3i-j). To determine whether three-photon longitudinal in vivo imaging affects the dynamics of oligodendrogenesis in the healthy brain, we implanted chronic cranial windows over the posterior parietal cortex in two groups of age-matched mice and compared the rate of oligodendrogenesis following longitudinal two-photon or three-photon imaging in the superficial cortex (0 – 350 μm depth) (Fig. 3k). We found no difference in the rate or total oligodendrocyte population gain (%) between groups imaged with different modalities (two-photon = 1.7±0.2 vs. three-photon = 1.4±0.3% gained per week, Fig. 3l). Finally, we explored whether longitudinal three-photon imaging affected oligodendrocyte survival. Mature oligodendrocytes in the healthy brain are remarkably stable and seldomly undergo cell death even when exposed to longitudinal two-photon imaging conditions^19,25,39^. In the superficial cortex, we found 1/477 mature oligodendrocytes were lost in the two-photon imaging group compared to 2/340 oligodendrocytes during three-photon imaging (Fig. 3m). In regions below those accessible with longitudinal two-photon imaging (350 – 1000 μm depth), 0/3326 tracked oligodendrocytes were lost during 9.5 weeks of longitudinal three-photon imaging in healthy mice (Fig. 3m; n = 4-5 mice per imaging modality). These results show that quantitative measures of health and behavior of oligodendrocytes obtained from long-term in vivo three-photon imaging in age- and region-matched mice are indistinguishable from those acquired via longitudinal two-photon imaging.

Increased population expansion of gray matter oligodendrocytes

We used long-term three-photon in vivo imaging to track oligodendrogenesis across the entire depth of the posterior parietal cortex (gray matter, GM) and the subcortical corpus callosum (white matter, WM) over multiple months in healthy mice (Fig. 4a-d). Similar to past histological studies^3,4,40^, we found that the white matter generated approximately 3X the number of new oligodendrocytes per volume compared to the cortical gray matter (33.8±5.5 vs. 10.9±1.3 OLs / imaging volume; 3.9±0.5 vs. 1.2±0.2 OLs / volume / week, Fig. 4e). This difference in the density of newly-generated oligodendrocytes may be related to regional differences in oligodendrocyte precursor cell (OPC) density and proliferation^41,42^. To examine OPC proliferation in these regions we delivered EdU, a thymidine analog carrying an ethynyl group, (5 mg/kg, 2x per day i.p.) for one week (5 mg/kg, 2x per day i.p.) to P70 mice (identical to onset of three-photon in vivo imaging). Tissue collection and post-hoc immunohistochemistry for Platelet-Derived Growth Factor Receptor Alpha (PDGFRα), a ubiquitous marker of oligodendrocyte precursors, and EdU were performed to assess OPC density and proliferation (Extended Data Fig. 7). We found that the density of OPCs is ~1.4X higher in the white compared to the gray matter of the PPC (248.6±23.8 vs. 181.2±7.7 OPCs/mm^2^). Furthermore, we found that a ~3.5X higher percentage of OPCs were EDU-positive in the white matter compared to the gray matter (51.6±6.6 vs. 14.5±2.5%). Together, these results suggest that region-specific regulation of OPC proliferation and density may support the increased number of adult-generated oligodendrocytes in the white matter.

In vivo imaging permits a unique perspective to evaluate changes in cellular behavior at the population level. We utilized this long-term approach to determine how the population of mature oligodendrocytes in gray and white matter regions changed over 2.5 months in the adult brain (Fig. 4f, Methods). In contrast to regional differences in the densities of newly-generated oligodendrocytes (Fig. 4e), the oligodendrocyte population in the adult gray matter expands by a greater percentage than the white matter (Fig. 4f). The total extent of oligodendrocyte population gain was significantly elevated in the healthy gray vs. white matter ((New OLs / Initial OLs) * 100; 19.1±2.1% vs. 12.1±1.3%, Fig. 4g). To examine the temporal dynamics of oligodendrocyte population growth in the adult gray versus white matter, we used exponential mechanistic growth modeling to describe the cell population-based decline in adult growth response protein 1 (Egr1), Kallikrein-6 (Klk6 and Prostaglandin D1 Synthase (Ptgds). These markers represent the distinct subpopulations of mature oligodendrocytes, MOL1, MOL2/3, and MOL5/6, respectively, that differ with development, brain region, and in responses to injury^6-8^. MOL1-positive oligodendrocytes were predominantly detected in the gray matter, MOL2/3-oligodendrocytes in the deep cortex and white matter, and MOL5/6 oligodendrocytes in both the gray and white matter (Fig. 4k-m). We found no differences in the relative expression of these oligodendrocyte subtypes between P140 and P60 indicating that heterogeneity in these regions is largely established by P60 (Fig. 4n). The representation of oligodendrocyte subpopulations exhibited regional heterogeneity at P140, as the MOL1-positive oligodendrocyte population was significantly larger in the gray matter (17.2 ± 3.4 vs. 0.6 ± 0.2%), the MOL2/3-positive population larger in the white matter (1.8 ± 0.5 vs. 3.8 ± 0.8%), and the MOL5/6-positive population proportionally higher in the white matter (27.4 ± 3.4 vs. 46.4 ± 4.7%) (Fig. 4o). Analysis of mid-thoracic spinal cord tissue processed in parallel with the brain tissue revealed a similar pattern of spatial heterogeneity to past studies^7^ (Extended Data Fig. 7). These data show that there is marked regional oligodendrocyte heterogeneity in the posterior parietal cortex and subcortical white matter well into adulthood (P140).

Cuprizone-mediated oligodendrocyte population loss does not differ across gray and white matter

Histological studies of regional differences in the cuprizone-mediated demyelination model have produced varied results that differ due to dose, age, time of administration, route of delivery, sex, and/or mouse strain^44-47^. To examine changes to gray and white matter populations of oligodendrocytes before, during, and after oligodendrogenesis with age^43^ (Extended Data Fig. 8). Oligodendrogenesis rates were similar between regions at early time points, but significantly decreased in the white matter by 15 weeks of age (2.3±0.3 vs. 1.3±0.2% per week, Fig. 4h). The time of 50% population growth did not differ between the gray and white matter (26.8±2.2 vs. 24.3±4.4 days post-P70, respectively, Fig. 4i). Binning the raw growth rate data by two-week intervals revealed that the white matter experiences a larger age-dependent decrease in population growth than the gray matter (1.0±0.4 vs. 2.5±0.5, WM, Weeks 5-6 vs. Weeks 1-2; 0.8±0.2 vs. 2.5±0.5, WM, Weeks 9-10 vs. Weeks 1-2, Fig. 4j). Together, these data reveal that the gray matter experiences greater oligodendrocyte population expansion that remains elevated later into life than the white matter.

To examine whether regional differences in oligodendrocyte population growth were mirrored by transcriptional heterogeneity in the healthy adult mouse brain, we collected tissue from MOBP-EGFP mice at P60 and P140 for RNA in situ hybridization (ISH) of Early demyelination, we implanted 10-week-old MOBP-EGFP mice with chronic cranial windows over posterior parietal cortex and fed them a diet of 0.2% cuprizone chow for 3 weeks while performing long-term in vivo three-photon imaging over approximately ten weeks (Fig. 5a). Previous work from our lab and others, showed that longitudinal in vivo imaging allows for the visualization of the overlapping time courses of oligodendrocyte loss and regeneration and this cuprizone treatment protocol results in ~80-90% oligodendrocyte loss in the superficial primary motor or somatosensory cortex^14,19^. To determine the extent of gray and white matter captured with longitudinal in vivo three-photon imaging, we used the THG signal to quantify gray and white matter regions of posterior parietal cortex. Our three-photon imaging parameters permitted longitudinal tracking of individual oligodendrocytes throughout the entire gray matter and the dorsal ~90% of the white matter over the entire imaging time course (Fig. 5b). The analyzed proportion of the white matter did not differ between healthy and cuprizone groups (84±8.0% vs. 86.8±3.5%, Fig. 5c). Long-term three-photon imaging allowed for the simultaneous visualization of the overlapping time courses of cuprizone demyelination and regeneration in the GM and WM (Fig. 5d-g). Volumetric cell loss mirrored differences in the regional density of mature oligodendrocytes (Fig. 1), as the white matter lost significantly more oligodendrocytes per volume than the gray matter (157.4±37.8 vs. 36.8±6.0 OLs / imaging volume, Fig. 5h). Next, we examined the effects of cuprizone-mediated demyelination on the oligodendrocyte population dynamics in gray and white matter regions. We found that the population loss of oligodendrocytes was equivalent in the gray versus white matter (Fig. 5i). These data indicate that the probability of survival of individual oligodendrocytes is similar across the gray and white matter regions, as would be expected from delivery of a systemic oligodendrocyte toxin at a saturating dosage. Accordingly, we found no differences in the total oligodendrocyte population loss (75.3±6.3% vs. 75.6±7.4%), rate of oligodendrocyte population loss (9.0±0.7% vs. 9.2±1.0% loss per week), or the timing of the inflection point of loss between the gray and white matter (0.4±1.4 vs. 2.5±2.0 days post cuprizone, Fig. 5j-l). Within-groups analyses of binned rates of demyelination revealed that gray matter oligodendrocyte loss occurs slightly earlier than in the white matter, yet there were no between-groups differences in the rate of oligodendrocyte loss (9.6±1.8 vs. 3.7±0.8 % per week, GM, Weeks ^−^2 to 0 vs. Week ^−^3; 18.9±1.9 vs. 3.7±0.8 % per week, GM, Weeks 0 to 2 vs. Week ^−^3; 22.6±3.1 vs. 4.1±2.5 % per week, WM, Weeks 1 to 2 vs. Week ^−^3, Fig. 5m). RNA ISH experiments showed that the prevalence of MOL1, MOL2/3, and MOL5/6-positive oligodendrocytes decreased across both the gray and white matter at four days post-cuprizone cessation (Fig. 5n-q, see Supplementary Table 3 for detailed statistics). While both surviving and newly-generated oligodendrocytes are present four days following cuprizone cessation (Fig. 5d,f), we found that the majority of oligodendrocytes at this time point are not labeled with MOL1, MOL2/3, or MOL5/6 subpopulations markers. Despite the reduction in oligodendrocyte subpopulation markers, we found that gray and white matter heterogeneity was preserved for MOL1 and MOL2/3 but abolished for MOL5/6 (1.8±0.7 vs. 0.1%, MOL1; 0.1±0.05 vs. 1.2±0.2%, MOL2/3; 3.1±1.1 vs. 4.9±1.6%, MOL5/6). Together, these data show that cuprizone-mediated demyelination results in similar loss of oligodendrocyte populations and transcriptional heterogeneity across the posterior parietal cortex and underlying corpus callosum.

Enhanced regeneration of oligodendrocyte populations in white compared to gray matter

Since cuprizone treatment affects oligodendrocyte populations in gray and white matter equivalently (see Fig. 5i,q), we compared the regenerative capacity of gray and white matter regions following cuprizone-mediated demyelination. We analyzed the region-specific extent, rate, and timing of oligodendrogenesis following cuprizone treatment (Fig. 6a-d). Similar to previous fate-mapping studies^47^, we found that there were more newly-generated oligodendrocytes in the white matter compared to the gray matter following cuprizone administration for all mice analyzed (Fig. 6e, 123±47.3 vs. 13.3±2.0 OLs/imaging volume). Next, we examined the restoration of the oligodendrocyte populations in these brain regions. We found that, even when normalizing to the initial population density, the subcortical white matter showed enhanced replacement of lost oligodendrocytes over the 2.5-month imaging timeline compared to gray matter (Fig. 6f). Forty-five days following removal of cuprizone, the total oligodendrocyte replacement was 37.6±3.6% vs. 68.0±11.3% for gray versus white matter, respectively (Fig. 6g). To analyze the temporal dynamics of post-cuprizone regeneration, we used three-parameter logistic modeling as described previously^14,48^ (Extended Data Fig. 8). Analyzing the Gompertz growth curves of regeneration revealed that the mean rate of oligodendrocyte population replacement per week during remyelination was 5.6±0.6% versus 11.1±1.9% for gray versus white matter (Fig. 6h). Yet, the inflection point of oligodendrocyte population growth did not differ suggesting that the time course of the replacement response was similar in both regions (Fig. 6f,i). This effect was further supported by analyzing the binned oligodendrocyte growth data in which both the gray and white matter were elevated at the 3-4 week post-cuprizone time bin (8.2±1.1 vs. 3.0±0.5% per week, GM, Weeks 3-4 vs. Week 7; 18.1±5.1 vs. 4.5±1.3% per week, WM, Weeks 3-4 vs. Week 7) yet with a significantly higher peak in the white matter (18.1±5.1 vs. 8.2±1.1% per week, Fig. 6j). Finally, since the gray matter replaces 37.6% of lost oligodendrocytes compared to 68% replacement in the white matter, we estimated the time required to reach 100% replacement for each region. Assuming linear growth after 7 weeks post-cuprizone, we estimate that ~14 weeks would be required to replace the lost oligodendrocytes in the gray matter whereas the white matter would require ~7 weeks highlighting the reduced recovery rate of gray matter regenerative oligodendrogenesis compared to the white matter. When we examined recovery of oligodendrocyte transcriptional subpopulations via RNA ISH experiments at 7 weeks post-cuprizone cessation (age-matched to final imaging time point) we found that the MOL1 and MOL2 populations returned to healthy proportions in GM, while MOL5/6 remained decreased in both regions (−14.9±2.0%, GM; −22.6±4.2%, WM, Fig. 6n). Interestingly, the pattern of healthy gray versus white matter heterogeneity for these markers was nearly restored to healthy conditions (Fig. 6o), despite the decreased prevalence of MOL5/6 at this time point (GM vs. WM; 15.8±3.3 vs. 1.5±0.3%, MOL1; 1.2±0.4 vs. 4.1±1.2%, MOL2/3; 12.6±2.0 vs. 24.0±4.2%, MOL5/6). Together, these data show that the white matter restores the lost oligodendrocyte population more effectively than the gray matter, and bulk gray and white matter regions re-establish spatial distributions of oligodendrocytes subtypes to healthy levels following a demyelinating injury.

Layer-dependent regulation of oligodendrocyte growth, loss, and regeneration

Recent studies using longitudinal in vivo two-photon imaging revealed layer-specific differences in cortical oligodendrogenesis following motor learning^14^ and cuprizone-mediated demyelination^19^, yet these studies were limited to the superficial ~300 – 400 μm of cortex. We utilized the enhanced imaging depth of three-photon microscopy to assess layer-specific differences in oligodendrocyte population dynamics across all cortical layers and the underlying corpus callosum in both the healthy brain and following cuprizone-mediated demyelination. Based on the myelo-, neuronal, axonal, and synaptic anatomy of the adult posterior parietal cortex (Fig. 7a), we divided gray and white matter regions into layers 1-3 (superficial cortex), layer 4 (thalamic input layer to the PPC), layer 5-6 (deep cortex), and the subcortical corpus callosum (CC). We assessed layer-dependent differences in oligodendrocyte generation and transcriptional subpopulations across four experimental time points spanning P60 to P140 in healthy and cuprizone-treated mice (Fig. 7b, Extended Data Fig. 9). In healthy mice, we found that there was increased oligodendrogenesis in L4 compared to the CC, suggesting that enhanced oligodendrogenesis in this specific cortical microenvironment drives increased population growth in the gray versus the white matter (2.6±0.2 vs. 1.4±0.2% per week, Fig. 7c-d, Fig. 3). Analyses of layer-specific oligodendrocyte subpopulations showed the following patterns in the healthy posterior parietal The proportion of MOL1-positive oligodendrocytes declined with cortical depth and were largely absent from the white matter (Fig. 7e). The proportion of MOL2/3-positive oligodendrocytes increased with depth and, within the cortex, were only detected in L5-6 (Fig. 7f). The white matter had increased proportion of MOL5/6 oligodendrocytes compared to the gray matter (Fig. 7g, see Supplementary Table 3 for detailed statistics). We found that cuprizone-induced oligodendrocyte loss did not differ across cortical or subcortical layers (Fig. 7h,i). The proportion of MOL1-postive oligodendrocytes was increased in layers 1-3 compared to the corpus callosum (Fig. 7j), MOL2/3-positive oligodendrocytes were increased in the corpus callosum compared to all cortical layers (Fig. 7k), while the proportion of MOL5/6 oligodendrocytes did not differ across cortical and subcortical layers (Fig. 7l, see Supplementary Table 3 for detailed statistics). When examining oligodendrocyte regeneration seven weeks following cuprizone removal, we found that the deep cortex (L5-6) had a decreased rate of oligodendrocyte replacement compared to the corpus callosum (5.6±0.56 vs. 11.2±1.9% per week, Fig. 7m,n). At 7 weeks post-cuprizone, the proportion of MOL1-postive oligodendrocytes was increased in layer 4 compared to the corpus callosum (Fig. 7o), MOL2/3-positive oligodendrocytes were increased in layers 5/6 and the corpus callosum compared to layers 1-3 and 4 (Fig. 7p), while the proportion of MOL5/6 oligodendrocytes was decreased compared to the corpus callosum (10.6±2.2 vs. 24.0±4.2%; Fig. 7q, see Supplementary Table 3 for detailed statistics).

Insufficient oligodendrogenesis and regeneration of MOL5/6 subpopulation in deep cortical layers

To determine the layer-specific capacity to generate heterogeneous oligodendrocyte subpopulations across health and disease, we examined the dynamics of oligodendrogenesis and transcriptional oligodendrocyte subpopulations in healthy and cuprizone treated mice. Compared to the healthy brain, we found that regenerative oligodendrogenesis following cuprizone treatment was increased by an order of magnitude (1 – 7% vs. 0.1 – 0.65%, Fig. 7d,n). Therefore, we normalized the modeled growth curves to the maximum value of oligodendrocyte gain or replacement to enable between-groups comparisons of healthy vs. regenerative oligodendrogenesis (Fig. 8a; Methods). We found that the scaled total oligodendrocyte population growth (0.3±0.03 vs. 0.5±0.1, Fig. 8a,b) and rate of growth (0.3±0.03 vs. 0.6±0.1, Fig. 7a,c) in L5–6 was decreased compared to healthy conditions, yet the growth curves had equivalent relative inflection points (Fig. 8d). Further analyses of the dynamics of regeneration revealed layer-dependent differences in response duration (full-width at half-maximum) and a decrease in the integrated response (area under the curve, above healthy) in L5-6 compared to the corpus callosum (19.89±2.14 vs. 11.70±2.37 days; 22.8±2.9 vs. 56.2±12.1%, Extended Data Figure 10).

Next, we analyzed the loss and restoration of layer-specific oligodendrocyte heterogeneity four days and seven weeks following cuprizone removal compared to age-matched healthy controls (Fig. 8e). To calculate the overall oligodendrocyte heterogeneity per layer, we determined the number of oligodendrocyte subpopulations that are present within each layer. If the proportion of oligodendrocyte subpopulations was higher than the minimum proportion in healthy control animals, this oligodendrocyte subpopulation was considered to be present within that layer. Four days following cuprizone removal, we found that oligodendrocyte heterogeneity was reduced in all cortical layers whereas oligodendrocyte heterogeneity in the corpus callosum remain unchanged (Fig. 8e). Seven weeks following cuprizone removal, we found that oligodendrocyte heterogeneity across all layers was restored to healthy levels (Fig. 8e). Following seven weeks of oligodendrocyte regeneration, we found that layer-specific MOL1- and MOL2/3-positive subpopulations were indistinguishable from healthy controls (Fig. 8f,g). However, while MOL5/6 oligodendrocytes were regenerated in layers 1-3 and 4, we found this subpopulation remained decreased in layers 5-6 and the white matter compared to healthy conditions (10.6±2.2% vs. 31.3±3.1% , L5-6; 24.0±4.2% vs. 46.4±4.7%, CC; Fig. 8h). In the corpus callosum MOL5/6 oligodendrocytes were increased at seven weeks compared to four days following cuprizone removal (24.0±4.2% vs. 4.9±1.6%) indicating ongoing, albeit incomplete, regeneration of this subpopulation. In contrast, the proportion of MOL5/6 oligodendrocytes in layers 5/6 did not differ at seven weeks compared to four days post-cuprizone (10.6±2.2% vs. 1.9±0.8%) indicating a deficit in MOL5/6 regeneration in layers 5/6 following cuprizone-mediate demyelination. Together these results show that deep cortical layers 5-6 have pronounced deficits in regenerative oligodendrogenesis and restoration of the MOL5/6-positive oligodendrocyte subpopulation compared to other cortical and subcortical layers.

Discussion

The generation of myelin continues throughout life via adult oligodendrogenesis and plays essential roles in cognition and tissue regeneration following injury or disease. Exploring the regulation oligodendrogenesis in specific brain regions is essential to understand how these processes may be harnessed to promote learning or recovery in the adult CNS. In this study, we empirically determined excitation and scanning parameters to allow for longitudinal in vivo three-photon imaging over multiple months without cellular reactivity or tissue damage. This approach enabled the analysis of oligodendrocyte generation and population dynamics in the healthy and demyelinated posterior parietal cortex and the subcortical corpus callosum at the level of individual oligodendrocytes. Long-term in vivo three-photon imaging confirmed previous findings via fate-mapping and EdU-labeling that the white matter generates substantially more new oligodendrocytes per brain volume compared to gray matter. Conversely, measuring the proportional change of adult gray and white matter oligodendrocyte populations over time revealed an elevated rate of population expansion in the gray matter in five-month-old adult mice. Following demyelinating injury in the absence of autoimmunity, we found that the white matter had enhanced replacement of lost oligodendrocytes irrespective of baseline regional differences in oligodendrocyte density in white and gray matter regions. Additional layer- and subregion-specific analyses revealed a deficiency in oligodendrocyte regeneration in the deep gray matter regions that correlated with a reduced recovery of the MOL5/6 oligodendrocyte subpopulation. As deep cortical layers are essential for transmitting cortical output signals, our results suggest that regional differences in oligodendrocyte regeneration may contribute to deficits in cognitive function observed in human MS patients. Overall, this work provides a roadmap for long-term in vivo three-photon imaging of cellular behaviors across the gray and white matter regions of the adult mouse brain.

Despite progress in deep tissue imaging, state-of-the-art two-photon (2P) microscopy systems have limited imaging depths in intact brain tissue and imaging depths decrease over longitudinal imaging timelines. These approaches have not been able to explore deeper areas of the adult rodent brain without tissue removal or insertion of optical lenses which cause widespread tissue damage and glial reactivity. Recent three-photon imaging studies using 1300 nm excitation to examine fluorescent transgenic mice established a practical maximum in vivo imaging depth of ~1.2 mm below the brain surface for acute imaging time points immediately following a craniotomy^22,49^. However, the high peak intensity of excitation pulses required for nonlinear three-photon excitation induce cumulative tissue damage when applied longitudinally over multiple months (Extended Data Fig. 5). In this study, we achieved longitudinal imaging depths of ~1 mm via optimization of optical and scanning parameters in chronic imaging windows from 3 weeks to over 3.2 months following a craniotomy (Figs. 1,3; Extended Data Fig. 5). As the effective attenuation length (EAL) of gray matter is nearly 2X that of the white matter^49^, slightly larger imaging depths may be achieved in regions of thicker neocortex, which do not require imaging through the highly scattering subcortical white matter. Our data presents guidelines to develop effective long-term three-photon imaging parameters for a variety of tissues without disrupting sensitive cellular behaviors or inducing cellular reactivity (Fig. 2; Extended Data Fig. 5). Future work exploring potential laser-induced phototoxicity on transcriptomic or proteomic changes following long-term three-photon imaging may also provide additional insights into the effects of these imaging modalities. Overall, the ability to make longitudinal measurements over weeks-months timescales via in vivo three-photon imaging delivers a powerful approach to study cellular behaviors across the gray and white matter regions of the adult mouse brain.

One advantage of longitudinal measurements of cellular behaviors is that the initial population(s) of labeled cells are defined at before, during, and after interventions. This enables the analysis of population growth or decline within a defined subregion with single-cell resolution. We used these methods to examine the link between regional changes in oligodendrogenesis and oligodendrocyte population dynamics. Across development and into adulthood, the rate of oligodendrogenesis in the mouse CNS slows asynchronously across CNS regions and undergoes population decline at late stages of life^4,50^. Although the population density of oligodendrocytes is significantly lower in the gray matter versus the white matter, we found that the population grew at a higher rate in the cortex than in the corpus callosum. These data imply that regional microenvironments regulate the carrying capacity of the oligodendrocyte population to maintain heterogenous myelination dynamics throughout life. Mature oligodendrocytes cluster into at least six specific transcriptional subpopulations in the healthy brain^6^ that have defined spatial preferences in the CNS^7,8^, yet it is unknown how specific regional microenvironments regulate this spatial and transcriptional heterogeneity. Complimentary cross-sectional analyses using RNA in situ hybridization revealed gray and white matter regional differences in the proportions of MOL1, MOL2/3, and MOL5/6 oligodendrocytes in the posterior parietal cortex that were stable across young adulthood (P60 to P140). Whether these differences in the transcriptional makeup of oligodendrocytes in these regions reflect specific cellular functions, or simply differences in the maturation state of these populations remains unknown. However, it is thought that the MOL5/6 subpopulation emerges with age in postnatal development and may represent the most-mature oligodendrocyte state^6^, supporting our findings that MOL5/6 is enriched in the white matter and that that the overall population growth declines more rapidly in the white compared to the gray matter. How these region-specific oligodendrocyte population growth dynamics are governed remains an open question. Microenvironmental cues may act at specific stages of oligodendrocyte maturation, such as OPC proliferation, differentiation, or premyelinating cell survival and integration into axonal circuits^25,51^. Similar to previous studies, we found that OPC proliferation is significantly higher in the subcortical white matter compared to the posterior parietal cortex^3,52^(Extended Data Fig. 7). As proliferation is linked to OPC differentiation or death^53^, these results suggest a reduced time to cell cycle exit in the white matter that maintains homeostatic population density in response to the increased frequency of oligodendrogenesis in this region. Alternatively, regional variation in integration rates of premyelinating oligodendrocytes^25^ in different microenvironments may regulate regional oligodendrocyte population dynamics. The recent discovery that specific red fluorescent proteins can be excited in the 1300 nm spectral window with high efficiencies^25,54,55^, opens the door to future work with dual-color, oligodendrocyte stage-specific genetic reporter approaches^56^ to dissect how regulation of specific oligodendrocyte maturational stages leads to variable population growth across cortical and subcortical regions.

Similar to lesions in MS patients, the cuprizone-mediated demyelination model results in overlapping oligodendrocyte death and regeneration^57^. The demyelinating injury caused by cuprizone treatment occurs in the absence of notable peripheral immune response^58^ making it an ideal model to evaluate the intrinsic capacity of oligodendrocyte regeneration without the confound of autoimmunity. We found that the survival rate of individual oligodendrocytes is similar across the gray and white matter regions as would be expected from a saturating concentration of an oligodendrocyte toxin. However, a number of factors can affect feeding behavior^59^ and varied oral consumption of cuprizone diet could directly influence the dosage and resulting extent of demyelination in a region-dependent manner at sub-saturating doses. Long-term in vivo imaging approaches are essential for analyzing the dynamics of the overlapping periods of oligodendrocyte loss and gain observed with cuprizone-mediated demyelination. Here, we showed that systemic cuprizone administration affects gray and white matter regions similarly, both in terms of oligodendrocyte cell death and changes in transcriptional heterogeneity, yet the white matter has an enhanced intrinsic capacity for regeneration of oligodendrocytes and their subpopulations compared to the gray matter. These gray and white matter differences in intrinsic capacity for regeneration highlight the need for additional insights into the regional effects of “remyelination therapies” (approaches that enhance oligodendrocyte regeneration^60^) to harness their full potential to increase remyelination and accelerate functional recovery following demyelinating injuries. Might remyelination therapies be more successful in augmenting regeneration in the gray matter, where endogenous regeneration is stunted? Further work examining multicellular interactions in regional microenvironments will aid in our mechanistic understanding of differences in regional oligodendrocyte regeneration. For example, the type 1 transmembrane neuropilin-1 is specifically expressed on white matter microglia and transactivates platelet-derived growth factor-α to induce OPC proliferation following demyelination^61^. Conversely, this increased white matter regeneration could be mediated by the recruitment of Gli1-positive adult neural stem cells from the subventricular zone^62^. Protein–protein interaction databases and recent advances in RNA sequencing technologies will likely provide key access to ligand–receptor pairs^63^ that can help explore the role of multicellular interactions in oligodendrocyte population growth and regeneration.

In MS patients, in contrast to our findings with cuprizone-mediated demyelination, lesions in the white matter regions are thought to have more severe demyelination and decreased remyelination compared to gray matter lesions^64,65^. MS white matter lesions differ from gray matter lesions by often displaying blood-brain barrier disruption, infiltrating peripheral immune cells, and complement deposition (reviewed by Geurts and Barkhof^66^) suggesting that the inflammatory microenvironment of white matter limits the regenerative capacity of oligodendrocyte populations. Indeed, mouse white matter oligodendrocyte precursor cells have delayed differentiation when exposed to a combination of interferon-γ and tumor necrosis factor-α compared to gray matter-derived OPCs in vitro^67^. Recent findings show that OPCs can cross-present antigen and upregulate the immunoproteasome subunit PSMB8 specifically in the white matter^10,68^ suggesting that oligodendrocyte regeneration is disproportionately affected in the inflammatory context of white compared to gray matter regions in MS patients. Furthermore, regional heterogeneity of oligodendrocytes and other cell types may regulate the disease progression and regenerative capacity in a region-specific manner^11^. Future work using long-term in vivo three-photon imaging of oligodendrocytes in gray and white matter regions of immune-mediated demyelination models will provide important insights into the role of inflammation in regional regenerative capacity of oligodendrocyte populations.

Recent longitudinal in vivo two-photon imaging following cuprizone-mediated demyelination showed that oligodendrocyte regeneration declined with depth across the superficial 300 μm of somatosensory cortex^19^. Our results confirm and extend these findings as we show that deep cortical layers 5-6 have reduced oligodendrocyte regeneration compared to the baseline level of healthy oligodendrogenesis. Additionally, we identified a specific oligodendrocyte subpopulation in deep layers 5-6 that fails to be reestablished following cuprizone-mediated demyelination. Since mature oligodendrocyte density increases with depth (Fig. 1d), one possibility is that increased myelin debris suppresses intrinsic mechanisms of oligodendrocyte regeneration and reestablishment of subpopulation heterogeneity in the deep cortex, but not in the white matter^69^. The consequences of regional deficits in oligodendrocyte regeneration and remyelination are not well understood. Cortical layers 5-6 contain the primary output neurons that project to downstream brain regions to control behavior^70^ and reduced oligodendrocyte regeneration in these regions could underlie the debilitating behavioral deficits experienced by MS patients such as impaired hand function^71^. Oligodendrocytes that closely appose layer 5 pyramidal neurons (satellite oligodendrocytes) locally shape their intrinsic excitability via buffering potassium through the glial syncytium^72^. Loss of these oligodendrocytes could underlie increases in neural firing rates resulting in hyperexcitability following demyelination as well as lead to a loss of local neural circuit synchronization^14,73^. Furthermore, regional variation in oligodendrocyte and myelin loss may lead to cell-type-specific neuron vulnerability and glial activation patterns relevant to neurodegeneration^74^. Remyelination is protective for damaged axons and recent work shows that cell-type specific remyelination is driven by a combination of cell type and axonal diameter^20^. Additional work to understand the specificity of myelin placement on neural circuits across gray and white matter in the healthy and diseased brain will greatly aid in uncovering the potential for remyelination to restore circuit and behavioral function following demyelination.

Emerging optical techniques and approaches will continue to reduce barriers for future work examining the long-term dynamics of subcellular structures such as individual myelin sheaths, axonal boutons, and dendritic spines in deep brain regions. For example, the three-photon effective attenuation length (EAL) is highly dependent on the inhomogeneity factor of the fluorescent signal, therefore adopting a sparse labeling approach will reduce out-of-focus background and increase SBR at depth^75^. In the current study, the MOBP-EGFP mice are advantageous for the study of oligodendrogenesis as all of mature oligodendrocytes and myelin sheaths are labeled^25^, however, this transgenic mouse line has less EGFP expression compared to previously lines used for three-photon microscopy (e.g. Thy1-EGFP^76^). Moreover, the broad distribution of EGFP signal throughout the cortex generated substantial out-of-focus background, which likely limited the SBR when imaging deep in the brain. Therefore, longitudinal quantifications of cell behavior in the current study were limited to the detection of cell bodies rather than subcellular morphological dynamics or single myelin sheaths. Sparse labeling approaches^25,77^, combined with enhanced detection techniques^22,78^ will allow for the application of longitudinal three-photon microscopy to track subcellular structures at even greater depths in biological tissues^79^. Innovations in adaptive optics like direct wavefront sensing^80^ and zonal aberration correction^33^ will further advance the application of three-photon microscopy to studies of biological dynamics within single cells. Overall, strategies that enhance SBR without increasing average excitation power will be paramount to push the limits of deep-brain three-photon imaging to uncover fundamental discoveries in neurobiology.

Methods

Animals

All animal experiments were conducted in accordance with protocols approved by the Animal Care and Use Committee at the University of Colorado Anschutz Medical Campus. Approximately equivalent numbers of male and female mice were used in these experiments and were kept on a 14-h light–10-h dark schedule with ad libitum access to food and water. Mice were housed at a temperature of 72 ± 2°C with 50 ± 20% humidity levels. All mice were randomly assigned to conditions and were precisely age-matched (±5 days) across experimental groups. C57BL/6N MOBP–EGFP (MGI:4847238) mice, which have been previously described^81^, were used for in vivo multiphoton imaging.

Custom three-photon microscope

The custom microscope consisted of a VIVO Multiphoton Open (3i) microscope, based on a Sutter Moveable Objective Microscope (MOM) stand, that was modified for multichannel three-photon imaging. The laser output from a regenerative amplifier with 1030 nm center wavelength, 70 W average power, <300 fs pulse duration, and adjustable repetition rate up to 2 MHz (Spirit-1030-70, Spectra Physics, 1 μJ pulse energy at 2 MHz) was wavelength converted to 1300 nm by a noncollinear optical parametric amplifier (NOPA-VIS-IR, Spectra Physics). The laser was operated at a repetition rate of 1 MHz (pulse picking = 2) and the final output power from the idler at this repetition rate was 0.8 to 1.1 W at 1300 nm. The power was modulated using a motorized rotation mount (KPRM1E/M - Ø1, Thorlabs) with a half-wave plate and Glan-Thompson prism. Beam conditioning of the NOPA output consisted of an expansion/collimating lens relay (f1 ¼ 75 mm, f2 ¼ 100 mm, Newport). The beam was expanded using a 4× reflective beam expansion telescope (BE04R, Thorlabs) to uniformly fill the deformable mirror (DM, ALPAO) and then demagnified with a telescope relay back into the microscope (f1 ¼ 500 mm, f2 ¼ 200 mm, Edmund Optics). A dual prism compressor system consisting of two 25 mm SF10 prisms cut at Brewster’s angle (10NSF10, Newport) and a gold roof mirror (HRS1015-M01, Thorlabs) was used to compensate for group delay dispersion. The beam was directed to the galvanometers (Cambridge Technologies) and through a scan lens (SL50-3P, Thorlabs), tube lens, and a 760 nm long-pass primary dichroic (Semrock). Three-photon excitation was focused through a high-NA multiphoton objective (XLPLN25XWMP2, 25x/1.05 NA, Olympus) with the correction collar set for the 0.15 mm coverslip at the surface of the sample. The fluorescent emission was separated from the excitation path by a long pass dichroic mirror, spectrally filtered (green channel 525 / 50 nm, THG channel 430 / 25 nm), and detected by photomultiplier tubes (H10770PA-40, Hamamatsu). The electronic signal was amplified, low-pass filtered and digitized. Laser-clocked acquisition was not implemented on the current system and 2 μs pixel dwell times were used for all experiments to allow for the collection of fluorescence from ~2 laser pulses per pixel. Data were acquired with SlideBook 2022 (Intelligent Imaging Innovations).

Cranial window surgery

Chronic cranial windows were prepared as previously described^53^ with minor modifications for three-photon imaging (described below). Six- to eight-week-old mice were anesthetized with isoflurane inhalation (induction, 5%; maintenance, 1.5–2.0%, mixed with 0.5 liter per min O2) and kept at 37 °C body temperature with a thermostat-controlled heating plate. After removal of the skin over the right cerebral hemisphere, the skull was cleaned and a 2 × 2 mm region of skull centered over posterior parietal cortex (1–3 mm posterior to bregma and 1–3 mm lateral) was removed using a high-speed dental drill. A piece of cover glass (VWR, No. 1) was then placed in the craniotomy and sealed with Vetbond (3M) and then dental cement (C&B Metabond). For head stabilization, a custom metal plate with a central hole of diameter = 6 mm was attached to the skull parallel to the implanted glass window, and the dental cement was drilled out around the perimeter of the window to allow objective advancement deep into the brain without headbar contact. A 5 mg / kg dose of carprofen (Zoetis) was subcutaneously administered before awakening and for two additional days for analgesia. In vivo imaging sessions began 2–3 weeks post-surgery and took place once every 4 days or weekly (see imaging timeline in Fig. 5). During imaging sessions, mice were anesthetized with isoflurane and immobilized by attaching the head plate to a custom stage (See Longitudinal three-photon imaging with adaptive optics section). The internal temperature of the mice was maintained at 37 ± 0.5 C using a feedback temperature controller (TC-1000, CWE) and the Isoflurane was scavenged using a Minivac Gas Evacuation Unit and Charcoal Fluosorber (Harvard Apparatus).

Longitudinal two-photon imaging

Chronic cranial windows were prepared over the PPC and in vivo imaging was performed as above. 4D two-photon datasets were acquired using a Bruker Ultima Investigator upright microscope equipped with Hamamatsu GaAsP detectors and a mode-locked Ti:sapphire laser (Coherent Ultra) tuned to 920 nm. The average power at the sample was between 5 - 30 mW, corresponding to an exponential power curve of 5 - 15 mW to 80 - 120 mW with depth. At least three MOBP-EGFP-labeled mature oligodendrocytes distributed across the field of view near the surface of the brain were used to identify the same cortical area over longitudinal imaging sessions. Image stacks were acquired with Zeiss W plan-apochromat ×20/1.0 NA water immersion objective giving a volume of 444 x 444 x 350 μm (X,Y,Z; 1024 x 1024 pixels) from the cortical surface.

Long-term three-photon imaging with adaptive optics

The custom stage for three-photon imaging consisted of 3D-printed headbar clamps for anesthetized imaging affixed to a small breadboard (MB2020/M, Thorlabs) mounted on 2-axis brass goniometers (GOH-60B50, Optosigma) with the rotation center height approximately level with the cranial window. The first day of imaging (2.5-3 weeks post-surgery) consisted of the following The window was flattened to be orthogonal to the excitation angle using the THG signal generated at the surface of the cranial window^33,82,83^. The tip/tilt of the stage goniometers were manually adjusted while maintaining the focal plane at the surface of the window until the field uniformity of the THG signal was maximized and did not translate significantly when moving in z. Imaging fields were chosen based on image quality at the superficial ~20 μm of the corpus callosum (~800 μm depth). Epifluorescent landmark images were taken at the end of the first time point to relocate the imaging field on subsequent time points. To correct for optical aberrations induced by the inhomogeneity of mouse brain tissue, we adopted an indirect, image-based correction technique that has been described previously^22,84^. The deformable mirror (DM, DM97-15, ALPAO) was controlled in SlideBook 2022 and custom MATLAB scripts were used to perform plotting and peak-finding for individual Zernike modes. AO correction was performed on the THG signal of the vasculature and transverse myelin sheaths in the deep cortex just above the corpus callosum (~750 μm depth) at 256 x 256 resolution, corresponding to 5.1 frames / second. To calibrate for spherical-defocus coupling, we used two-photon imaging of a small ROI surrounding a single 0.2 μm fluorescent bead in a 2-dimensional preparation. The peak intensity of the bead sample was plotted against the spherical and defocus amplitudes at each z-position and the linear relationship between the modal amplitudes was applied to the DM during correction. The raw optimization data was reviewed to confirm that the focal plane was constant at different AO configurations. Parameters for AO correction consisted of the number and order of Zernike modes for correction, DM stroke range, and the stroke increment. We corrected first for the spherical mode followed by astigmatism, coma, trefoil, and then higher-order modes with a DM range of −2 to 2 um and an increment of 0.3, which resulted in the acquisition of 196 images in ~50 seconds (including analysis time). Zernike mode settings were optimized based on the mean intensity of the THG signal, which resulted in the greatest improvements in EGFP-positive oligodendrocyte signal to background (SBR) ratio in the subcortical WM (Extended Data Fig. 3). A single AO iteration was performed for chronic structural imaging in healthy mice. The single-plane AO correction deep in the cortex resulted in a suboptimal DM pattern for signal quality in the superficial cortex of the z-column. Therefore we implemented a dynamic linear spherical aberration correction that was updated with imaging depth without the need to reoptimize at each z plane. After optimizing the DM settings the power map was set stepwise starting from the meninges. 11 power map points were set (0, 50, 100-1000 μm at 100 μm intervals) using the fluorescent counts on the THG detection channel. The power was slowly increased at 0.2 (surface) or 1% intervals while monitoring the fluorescence intensity histogram to obtain a dynamic range of signal intensity between 0-7000 counts without saturation. In cuprizone mice, the power modulation maps were reduced by ~10% during the two weeks surrounding peak demyelination (16 – 32 days post-cuprizone) to account for decreased excitation scattering after myelin loss. Large-volume z-stacks were acquired with bidirectional scanning at 512 x 512 pixels (385 x 385 μm), dwell time = 2 μs, frame averaging = 2, and z-step = 3 μm, and scan time ~.8 frames / s. The PMT gain was set to 91% for THG and 93% for GFP and a custom blackout box was built to reduce ambient light noise. Raster scanned images are only recorded during ~80% of the x-galvonometer sweep (i.e. during constant velocity), therefore the Spirit pulsed output was blanked during the overscan to reduce the average power applied to the tissue. Finally we implemented a one minute frame pause after every three minutes of z-scanning to allow for periodic heat dissipation. For subsequent imaging time points, mature oligodendrocytes in the superficial cortex were aligned to the first time point using ROI overlays around the cell body positions on day one, the spherical depth correction was applied to the DM as in day one, and the AO correction was performed daily at the same depth before acquisition.

Cuprizone-mediated demyelination

Global demyelination was induced in our C57\B6N MOBP–EGFP mice using 0.2% cuprizone (Sigma Chemical, C9012), stored in a glass desiccator at 4 °C. Cuprizone was added to powdered chow (Harlan), mixed for ~10 minutes to ensure homogeneity, and provided to mice in custom feeders (designed to minimize exposure to moisture) for 3 weeks on an ad libitum basis. Feeders were refilled every 2–3 days, and fresh cuprizone chow was prepared weekly. Healthy mice received normal powdered chow in identical feeders. Powdered chow was introduced 2-3 days before cuprizone to ensure acclimation to the powdered food. Cages were changed weekly to avoid build-up of cuprizone chow in bedding, and to minimize reuptake of cuprizone chow following cessation of diet via coprophagia. This 3-week demyelination model that has been previously described^85^, results in ~88% loss of MOBP-EGFP oligodendrocytes in the superficial primary motor cortex, and allows for in vivo tracking of the overlapping time courses of demyelination and oligodendrocyte regeneration. Cuprizone mice were weighed weekly during and after cuprizone feeding to confirm consumption of the drug.

Image processing and analysis

4D longitudinal image stacks were analyzed using Fiji^86^. Cortical layers were defined using the Allen Reference Atlas^87^ and in vivo THG imaging data. Layer depth was calculated for individual mice using the depth of the white matter THG signal according to the equation below to eliminate variability in mouse size, field of view location, and imaging

Z depth of subregion (Allen Reference Atlas)Z depth of total cortex (Allen Reference Atlas)×Z depth of total cortex(invivo)

Unprocessed 1000 μm 4D stacks were cut into substacks of layers 1-3 (0 to 356 +/−21 μm), layers 4-6 (469 +/−22 μm) and the subcortical WM (96 +/−6 μm). Image registration was performed iteratively with the Poorman3Dreg plugin for lateral registration followed by the Correct3D drift plugin^88^ (EGFP channel, rigid body registration, 30x30x20 pixels X/Y/Z). Additional iterations of Correct3D drift were performed on high-signal ROIs as needed. The registered datasets were further processed with a median filter to remove salt and pepper noise and individual oligodendrocytes were manually tracked using custom FIJI scripts as previously reported^10^. Images generated for figures were brightness and contrast corrected for clarity and maximum z-projection widths are defined in the figure legends. Stable, new, and lost cells were segmented and pseudo-colored using the Lasso tool in Photoshop CS6 (Adobe).

Analyses of imaging performance and white matter fiber orientation

For two-photon versus three-photon imaging comparisons, oligodendrocyte cell bodies were counted within x-y subregions greater than or equal to 200 x 200 μm^2^ and oligodendrocytes were only counted if SBR was greater than 2 (as measured by a line scan profile in ImageJ). For the SBR analyses, oligodendrocytes were randomly sampled throughout the imaging volume to include the full range of oligodendrocyte SBRs (i.e. dim, medium, and bright) with at least 10 oligodendrocytes / 100 μm z-depth. ROIs were drawn around the oligodendrocyte cell body and then immediately adjacent to the cell body in a background region. Single oligodendrocyte SBRs were calculated as the max intensity of the cell body / max intensity of the background. AO improvements in oligodendrocyte signal intensity were quantified in the same manner, while improvements in resolution were calculated on oligodendrocyte cell bodies or single myelin sheaths as the full width at half-maximum in the lateral or axial dimensions. Fiber orientation calculations were performed on single z-slices within the WM using the OrientationJ plugin^89,90^.

EdU labeling and detection

To quantify regional differences in OPC proliferation and oligodendrocyte generation, EdU (A10044, Invitrogen) was diluted in sterile saline solution (0.9% NaCl) and injected intraperitoneally from P70-P76 (matched to starting age of in vivo imaging timeline) twice daily at 5 mg per kg for a total of 7 days. The time course and 5 mg per kg doses were chosen based on previously published work^91,92^, with the goal of labeling a substantial number of proliferating OPCs in the gray and white matter without saturating the population within either region^93^ and without generating false positives due to EdU binding at sites of DNA repair^94^ or inducing DNA damage pathways^95^. EdU-labeled cells were visualized using the AlexaFluor-647 Click-iT EdU Cell Proliferation Kit for Imaging (C10340, Invitrogen). The Click-iT reaction was performed per the manufacturer specifications immediately following the blocking step in the immunofluorescence protocol.

Immunofluorescence analyses

Mice were anesthetized with an intraperitoneal injection of sodium pentobarbital (100 mg per kg body weight) and transcardially perfused with 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.0–7.4) immediately following the final imaging time point or 16-20 h post-exposure in the laser damage experiments. Brains were postfixed in 4% paraformaldehyde for 1-2 h at 4 °C, transferred to 30% sucrose solution in PBS (pH 7.4) and stored at 4 °C for at least 24 h. Brains were extracted, frozen in TissuePlus OCT, sectioned coronally at 30 μm, and immunostained as free-floating sections. Sections were incubated in blocking solution (5% normal donkey serum, 2% bovine γ-globulin, 0.3% Triton X-100 in PBS, pH 7.4) for 1–2 h at room temperature, then overnight at 4 °C in primary antibody. When staining for Aspartoacylase (ASPA) sections were incubated with Liberate Antibody Binding Solution (L.A.B., Polysciences) for 10 minutes prior to the blocking stop. Secondary antibody incubation was performed at room temperature for 1.5 h. Sections were mounted on slides with Vectashield antifade reagent (Vector Laboratories). Images were acquired using a laser-scanning confocal microscope (Leica SP8). For laser damage experiments, mild columnar damage regions (positive control tissue) were created using identical scanning settings as in the longitudinal group, but with the power map right-shifted to saturate the THG channel. 2-3 coronal sections of the PPC spaced ~60 μm apart were chosen based on the stereotaxic coordinates of the 385 x 385 μm in vivo imaging field and aligned to the reference atlas using anatomical landmarks. Bulk intensity measurements were made as previously described^31,96^ on a 1000 x 1000 μm ROI drawn directly under the 385 x 385 μm imaging field, and in the identical region on the contralateral hemisphere. Irregularly-shaped lesions were cropped with a minimum area of 600 x 600 μm. For cell counts, images with a 0.56 μm pixel size were thresholded using the maximum entropy filter and then cell bodies in individual channels were segmented using FIJI’s particle analysis tool^86^ and then proofread manually using the ROI manager. The settings for the particle analyzer were as Size 40 - 2000, Circularity: 0-.7, Microglia; Size 30 - 2000, Circularity: 0-1, Oligodendrocytes; Size 200 - 3500, Circularity: 0-1 Neurons; Size 30 - 2000, Circularity: 0-0.5; Astrocytes). For the OL-specific analyses of phototoxicity, the mean intensity of each segmented MOBP-EGFP ROI was measured together with a background ROI in the superficial cortex. Cells were considered HSP70/72- or γ–H2a.X-positive if the mean intensity within the cell body was greater than two times the background measurement. Automated segmentation of the vasculature was performed as previously described^37,38^. Briefly, one random section from each mouse was concatenated into a z-stack and ~15 blood vessels of varying diameters and background regions were traced for each image using the pixel classifier. The classifier was saved and batch-applied to all the images in the dataset (available upon request), which were then binarized and analyzed for percentage area of vascular coverage. Pericytes were counted as above. All confocal images shown in Fig. 3 and Extended Data Fig. 6 were acquired with identical power settings, processed with a 100 pixel rolling ball background subtraction and 0.7 pixel gaussian blur, and brightness/contrast correction was applied identically across channels.

RNA in situ hybridization with immunofluorescent detection of EGFP In situ hybridization (ISH) was performed using the commercially available RNAscope V2 kit by adapting the manufacturer’s protocol (ACD Biotechne 323100). This procedure was specifically modified to include antibody labeling of EGFP+ oligodendrocytes in the same tissue sections labeled with three mRNA probes. Probes used in this study were designed for mouse Ptgds-C1 (ACD Biotechne 492781), Klk6-C2 (ACD Biotechne 493751), and Egr2-C3 (ACD Biotechne 407871). Briefly, 20 um, PFA-fixed, tissue sections from brain and spinal cord were washed and mounted on Superfrost microscope slides. Tissue dehydration, antigen retrieval, protease treatment, and probe hybridization were carried out according to the manufacturer’s protocol. Amplification and HRP reaction steps for C2, C3, and C1 (in this order) were carried out without deviation from the manufacturer’s protocol. Briefly, slides were first washed with 1x RNA Scope wash buffer 2x for 2 minutes at room temperature before incubation with Amp 1 for 30 minutes at 40C. This process was then repeated for Amp 2 and Amp 3. Following amplification steps slides were washed 2x for 2 minutes each in RNAscope wash buffer before incubation with the respective HRP conjugate for 15 min at 40C. Next, sections were washed 2x for 2 minutes each and incubated in the corresponding fluorophore for 30 minutes at 40C (C2, TSA Vivid 570 1500 in TSA buffer, C3 TSA Vivid 650 2000 in TSA buffer, or C1 Opal TSA-DIG (Akoya) 750 in TSA buffer). After HRP reaction slides were washed 2x for 2 minutes, incubated in HRP blocker for 15 min at 40C, and finally washed 2x for 2 minutes. This was then repeated for the remaining 2 probes. Following the final HRP blockade, slides were incubated in blocking buffer (5%NDS, 0.3% Triton X-100 in PBS) for 60 min at room temperature. Slides were then incubated with chicken anti-GFP 250 (AVES, GFP-1020) for 48 hours at 4C. Slides were next washed 3x for 15 minutes in PBS. Immunohistochemistry was completed with incubation in Donkey anti-chicken 488 500 (Jackson, 703-545-155) for 60 minutes. ISH was completed by washing slides 3x for 15 minutes in PBS before incubation in Polaris 780 (1:125 in diluent buffer, Akoya) for 30 minutes at 40C. Finally, slides were washed 2x for 2 minutes in RNAscope wash buffer and counterstained with RNAscope DAPI for 30 seconds before cover slipping with Prolong Gold (Invitrogen, P36934) for imaging.

RNAscope image acquisition and processing

Fluorescent images used for RNAscope quantification were acquired using the Olympus DSU Spinning Disc microscope in the Advanced Light Microscopy Core at the University of Colorado Anschutz Medical Campus. To obtain a large field of view, adjacent images were acquired using a 20x air objective and a 10% tile overlap. A z-stack with 4 um steps was used to capture the entire thickness of each tissue section and reduce the detection of false positive labeled cells as described previously^7^. Images were stitched and max projected using a custom ImageJ macro based on the BaSiC plugin^97^.

RNAscope quantification

Following imaging, images were analyzed using Qupath Software 0.4.3^98^. RNA expression per cell was quantified using a custom analysis pipeline. Briefly, cortical layers were measured and drawn onto each tissue section as described above (Image Processing and Analysis) to allow for quantification of mRNA puncta within cells in each individual cortical layer. Nuclei were semi-automatically segmented based on the DAPI channel. Nuclear segmentations were dilated to encompass the predicted area of each cell. Finally, mRNA puncta were segmented and assigned to DAPI+/EGFP+ cell regions for oligodendrocyte-specific quantification. Threshold values for each channel were based on the background signal in each image. Cell and puncta detection was performed on each section in Qupath and the spreadsheets containing the number, size, and area of puncta were exported and processed to calculate the percentage of each population within each cortical layer using custom R scripts (https://github.com/sbudoff/OligoRNAscopeTools). We used thresholds of 14, 12, and 7 estimate puncta per GFP positive cell of Egr2+, Klk6+, and Ptgds+, respectively, to assign oligodendrocyte populations. For Klk6, we observed that positive cells in the cortical sections expressed high levels of entirely clustered puncta and therefore only included cells with 1 or 2 clusters that had an estimated puncta number greater than 12. To calculate the number of oligodendrocyte classes that are present within each layer we calculated a threshold which was the lowest proportion of cells present in healthy control animals. If an animal had a proportion of oligodendrocytes higher than this threshold it was considered to have this class present within that cortical layer. We used a threshold of 5.73%, 0.394%, and 5.82%, of MOL1, MOL2/3, and MOL5/6, respectively, to determine the presence or absence of each oligodendrocyte class.

Statistics and Reproducibility

Sample sizes were comparable to relevant publications^8,10,15,80^. All micrographs are representative images showing phenomena seen in multiple mice. All data were initially screened for outliers using IQR methods. All mice in a litter underwent cranial window surgery and concurrent two-photon imaging and training timelines were designated to be a “batch”. When possible, experimental groups were replicated in multiple batches with multiple experimental groups per batch. To analyze in vivo changes in oligodendrogenesis, we used cumulative population growth and decline measures as previously described^10^. Briefly, oligodendrocyte gain / loss (%) was calculated as the change in oligodendrocyte number / initial cell number * 100, the rate of oligodendrocyte gain/loss (%) was calculated as the % change in oligodendrocyte number divided by a time interval (e.g. Remyelination phase), and oligodendrocyte replacement (%) as total oligodendrocyte gain / total oligodendrocyte loss * 100. The oligodendrocyte replacement (%) metric is used to control for differences in the magnitude of demyelination, as oligodendrocyte gain is strongly correlated with oligodendrocyte loss in the cuprizone model^1081,82^. To analyze regional differences in the rate and timing of oligodendrogenesis, we used statistical modeling of cell growth and decline. As healthy oligodendrogenesis rates decline with age^99^, we found that this process was best-fit using the exponential mechanistic growth model that is commonly used to describe population growth with competition for resources^100^. Extended Data Fig. 8 shows the mechanistic growth curves for example mice and group means along with R^2^ values, which were greater than or equal to 0.91 (cortex) and 0.96 (CC) for all analyzed mice. The exponential mechanistic growth equation was calculated according to the equation

a∗(1−be(−c∗t)

Where a=Asymptote, b=Scale, c=Growth rate, and t=time (days). Cuprizone potently inhibits the survival of new oligodendrocytes when it is actively being administered in the diet and cuprizone-induced cumulative OL loss and gain curves are sigmoidal and can be effectively modeled using three-parameter logistic regression^85^. In this study, we used a three-parameter Gompertz model commonly used to describe tumor growth^101,102^ that was robust to asymmetrical differences in the timing of the inflection point. Extended Data Fig. 8 shows the Gompertz growth curves for example mice and group means along with R^2^ values, which were greater than or equal to 0.99 (cortex) and 0.99 (CC) for all analyzed mice. The Gompertz function was calculated according to the equation

f(t)=ae(−ebx(t−c))

Where a=asymptote, b=growth rate, c=inflection point, and t = time (days).

Because the goal of this study was to test the hypothesis that different brain regions are independent with respect to oligodendrogenesis, we used unpaired rather than paired statistics for single comparisons between gray and white matter regions (unpaired two-tailed t-test, parametric; Mann Whitney U test, nonparametric). Statistical comparisons between multiple brain regions were made using one-way ANOVA followed by Tukey’s HSD post-hoc test (parametric) or Kruskal-Wallis followed by Dunn test (nonparametric). To compare oligodendrogenesis in the healthy brain to cuprizone remyelination, we scaled the datasets to the maximum value of healthy or cuprizone oligodendrogenesis. To compare healthy vs. cuprizone-induced oligodendrogenesis metrics we used full-factorial 2-way ANOVA for the interaction effect followed by multiple comparisons with Bonferroni correction. Bonferroni-adjusted significance levels are indicated in the figure legends. Normality was assessed using the Shapiro-Wilk Test and outliers were assessed by visualizing the interquartile ranges. All statistical analyses and modeling were conducted in JMP 16 (SAS).

Extended Data

Supplementary Material