This protocol presents methods to characterize the neuroinflammatory and hemodynamic response to mild traumatic brain injury and to integrate these data as part of a multivariate systems analysis using partial least squares regression.
Mild traumatic brain injuries (mTBIs) are a significant public health problem. Repeated exposure to mTBI can lead to cumulative, long-lasting functional deficits. Numerous studies by our group and others have shown that mTBI stimulates cytokine expression and activates microglia, decreases cerebral blood flow and metabolism, and impairs cerebrovascular reactivity. Moreover, several works have reported an association between derangements in these neuroinflammatory and hemodynamic markers and cognitive impairments. Herein we detail methods to characterize the neuroinflammatory and hemodynamic tissue response to mTBI in mice. Specifically, we describe how to perform a weight-drop model of mTBI, how to longitudinally measure cerebral blood flow using a non-invasive optical technique called diffuse correlation spectroscopy, and how to perform a Luminex multiplexed immunoassay on brain tissue samples to quantify cytokines and immunomodulatory phospho-proteins (e.g., within the MAPK and NFκB pathways) that respond to and regulate activity of microglia and other neural immune cells. Finally, we detail how to integrate these data using a multivariate systems analysis approach to understand the relationships between all of these variables. Understanding the relationships between these physiologic and molecular variables will ultimately enable us to identify mechanisms responsible for mTBI.
Panoramica
Mild traumatic brain injuries (mTBIs) impact ~1.6-3.8 million athletes annually1. These injuries, including sub-concussive and concussive injuries, can leave patients with transient physical, emotional, psychological and cognitive symptoms2. Moreover, repetitive mTBI (rmTBI) sustained within a “window of vulnerability” can lead to cumulative severity and duration of cognitive consequences that last longer than the effects of a single mTBI alone3, and ultimately even to permanent loss of function4,5,6. Although many patients recover within a relatively short time frame (<1 week), 10-40% of patients suffer longer lasting effects of mTBI for > 1 month, with some lasting up to 1 year3,7,8,9. Despite the high prevalence and lasting consequences of these injuries, injury mechanisms are poorly understood and no effective treatment strategies exist.
Given the high variability in outcomes after mTBI/rmTBI, one challenge in identifying early-stage molecular triggers from tissue obtained in terminal mTBI/rmTBI studies is the lack of longitudinal data demonstrating definitive "acute molecular links" of these molecular triggers to longer-term outcomes. To overcome this challenge, our group has discovered that acutely reduced cerebral blood flow measured acutely using an optical tool called diffuse correlation spectroscopy (DCS), strongly correlates with longer-term cognitive outcome in a mouse model of rmTBI10. Using this hemodynamic biomarker, we showed that mice with acutely low cerebral blood flow (and, by extension, worse predicted long-term outcome) have concomitant acute increases in neuronal phospho-signaling within both MAPK and NFκB pathways, increases in neuronal expression of pro-inflammatory cytokines, and increases in expression of the phagocyte/microglial marker Iba111. These data suggest a possible role for neuronal phospho-signaling, cytokine expression, and microglial activation in both the acute regulation of cerebral blood flow post injury as well as in triggering a signaling cascade that leads to neuronal dysfunction and worse cognitive outcome. Herein, we detail our approach to simultaneously probe both the hemodynamic and neuroinflammatory environment after rmTBI and how to integrate these complex datasets. Specifically, we outline procedures for four key steps to this comprehensive approach: (1) a weight-drop model of mild traumatic brain injury, (2) assessment of cerebral blood flow with diffuse correlation spectroscopy, (3) quantification of the neuroinflammatory environment, and (4) data integration (Figure 1). Below, we provide a brief introduction to each of these key steps to help guide readers through the rationale behind our methods. The remainder of the manuscript provides a detailed protocol for each of these key steps.
Weight-drop model of mild traumatic brain injury
Although many excellent preclinical models of repetitive mild TBI exist12,13,14,15,16,17,18, we employ a well-established and clinically relevant weight-drop closed head injury model. Key features of this model include (1) blunt impact of the intact skull/scalp followed by unrestricted rotation of the head about the neck, (2) no overt structural brain injury, edema, blood–brain barrier damage, acute cell death, or chronic brain tissue loss, and (3) persistent (up to 1 year) cognitive deficits that emerge only after multiple hits19 (Figure 2).
Assessment of cerebral blood flow with diffuse correlation spectroscopy
Diffuse correlation spectroscopy (DCS) is a non-invasive optical technique that measures blood flow5,20,21. In DCS, a near-infrared light source is placed on the tissue surface. A detector is placed at a fixed distance from the source on the tissue surface to detect light that has multiply scattered through the tissue (Figure 3). Scattering off moving red blood cells causes the detected light intensity to fluctuate with time. A simple analytical model known as correlation diffusion theory is used to relate these intensity fluctuations to an index of blood flow (CBFi, Figure 4). Although the units of CBFi (cm2/s) are not the traditional units of flow (mL/min/100 g), a previous study in mice has shown that CBFi strongly correlates with cerebral blood flow measured by arterial spin labeled MRI21.
For reference, the DCS instrument used here was built in-house and is comprised of an 852 nm long coherence-length laser, an array of 4 photon counting avalanche photodiodes, and a hardware autocorrelator board (single tau, 8 channel, 100 ns minimum sample time)21,22. Data is acquired with homemade software written in LabView. The animal interface for the device consists of a 400 μm multimode source fiber (400-2200 nm wavelength range, pure silica core, TECS Hard Cladding) and a 780 nm single mode detector fiber (780-970 nm wavelength range, pure silica core, TECS Hard Cladding, 730 ± 30 nm second mode cut-off) spaced 6 mm apart and embedded in a black 3D-printed sensor (4 mm x 8 mm, Figure 3).
Quantification of the neuroinflammatory environment
Although neuroinflammation is regulated by diverse cellular processes, two key relevant mechanisms are extracellular signaling by cytokines/chemokines and intracellular signaling by phospho-proteins. To investigate the neuroinflammatory environment of the brain post-injury, brains are extracted from mice, microdissected, and cytokines/chemokines and phospho-proteins are quantified using Luminex (Figure 5, Figure 6, Figure 7). Luminex multiplexed immunoassays enable simultaneous quantification of a diverse collection of these proteins by coupling enzyme-linked immunosorbent assays (ELISAs) to fluorescently tagged magnetic beads. Distinct fluorescent tags are used for each protein of interest, and beads of each tag are functionalized with a capture antibody against that particular protein. Hundreds of beads for capturing each protein are mixed together, placed in a 96 well plate, and incubated with sample. After sample incubation, a magnet is used to trap the beads in the well while the sample is washed out. Next, biotinylated detection antibody binds to the analyte of interest to form an antibody-antigen sandwich similar to a traditional ELISA, but with the ELISA for each protein occurring on a different fluorescently tagged bead. Adding phycoerythrin-conjugated streptavidin (SAPE) completes each reaction. The Luminex instrument then reads the beads and separates the signal according to each fluorescent tag/protein.
Data integration
Because of the large number of analytes (e.g., cytokines) measured in the Luminex assay, data analysis can be difficult to interpret if each quantified protein is analyzed individually. To simplify analysis and to capture trends observed among analytes, we use a multivariate analysis method called partial least squares regression (PLSR, Figure 8)23. PLSR works by identifying an axis of weights corresponding to each measured protein (i.e., cytokines or phospho-proteins, referred to as “predictor variables”) that together optimally explain co-variance of the measured proteins with a response variable (e.g., cerebral blood flow). The weights are referred to as “loadings” and are assembled into a vector known as a latent variable (LV). By projecting (referred to as “scoring”) the measured protein data on each of two LVs, the data can be re-plotted in terms of these LVs. After computing the PLSR, we use a varimax rotation to identify a new LV that maximizes the covariance between the sample projections onto the LV and the predictor variable24. This approach allows us to define LV1 as the axis for which the variance of the response variable is best explained. LV2 maximizes co-variance between the response variable and LV1 residual data, which may be associated with biological or technical variability between samples. Lastly, we conduct a Leave One Out Cross Validation (LOOCV) to ensure that the PLSR model is not heavily dependent upon any one sample23.
In this protocol, we detail methods to characterize the neuroinflammatory and hemodynamic tissue response to mTBI. The general workflow is outlined in Figure 1. In this protocol, mice are subject to one or more mTBIs using a weight-drop closed head injury model. Cerebral blood flow is measured longitudinally before and at multiple time points after injury. At the time point of interest for interrogation of neuroinflammatory changes, the animal is euthanized, and the brain is extracted. Brain regions of interest are isolated via microdissection and then lysed to extract protein. Lysates are then used for both Luminex multiplexed immunoassays of cytokine and phospho-protein expression as well as Western blot. Finally, this holistic dataset is integrated using a partial least squares regression analysis.
All animal procedures are approved by Emory University Institutional Animal Care and Use Committee (IACUC) and followed the NIH Guidelines for the Care and Use of Laboratory Animals.
1. Weight-drop model of mild traumatic brain injury
2. Assessment of cerebral blood flow with diffuse correlation spectroscopy
3. Multiplexed quantification of cytokines and phospho-proteins using luminex assays
4. Partial least squares regression
NOTE: Sample R code and a sample data spreadsheet are provided to carry out the Partial Least Squares Analysis.
Previously collected data were taken from prior work in which a group of eight C57BL/6 mice were subjected to three closed-head injuries (Figure 2) spaced once daily11. In this work, cerebral blood flow was measured with diffuse correlation spectroscopy 4 h after the last injury (Figure 3, Figure 4). After post-injury CBF assessment, the animals were euthanized, and brain tissue was extracted for quantification of cytokines and phospho-proteins via immunoassay (Figure 5). We also quantified the phagocyte/microglial activation marker Iba1 via Western blot (methods described in11). Brain tissue from each mouse was lysed, and total protein concentration was measured using a BCA assay. Multiplexed cytokine quantification was conducted using the Milliplex MAP Mouse Cytokine/Chemokine 32-Plex, which was read using a Luminex MAGPIX system (Figure 6). A linear range analysis was conducted to determine an appropriate protein loading (12 µg of protein per 12.5 µL of lysate) (Figure 7) prior to collecting data from all samples.
Cytokine data was prepared for analysis by subtracting background measurements from sample data, and then z-scoring data for each analyte (Figure 8A). A heatmap was generated from z-scored data to visualize differences in cytokine expression among animals. Partial Least Squares Regression (PLSR) was conducted using the phagocyte/microglial activation marker Iba1 as the response variable and cytokine measurements as the predictor variables (Figure 8B). A varimax rotation was performed to maximize the co-variance of the data on LV1 with the Iba1 measurements (Figure 8D). High loading weights in LV1 (Figure 8C) correspond with the cytokine expression most associated with high expression of Iba1. Linear regressions between Iba1 and cytokines show that those cytokines with the greatest loading weights in LV1 were also statistically significant (Figure 8E).
Figure 1: Typical workflow. First, mice undergo a weight drop closed head injury, and then cerebral blood flow (CBF) is measured using diffuse correlation spectroscopy. Next, brains are collected, regions of interest are micro-dissected and snap frozen using liquid nitrogen. In preparation for the Luminex immunoassay, proteins are lysed, and total protein concentration is measured by bicinchoninic acid assay. Lysates are used for Western blot of proteins of interest and Luminex assays for cytokines and phospho-proteins. Data from CBF, Western blot, and Luminex are integrated using partial least squares regression (PLSR). Please click here to view a larger version of this figure.
Figure 2: Closed-head weight drop model of mild traumatic brain injury. (A) The anesthetized mouse is grasped by the tail and placed on a taut sacrificial membrane underneath a guide tube. A 54 g weight is dropped from 1 m onto the dorsal aspect of the head. (B) Within ~1 ms post-impact, the mouse’s head has rapidly rotated about the neck as it breaks through the sacrificial membrane. (C) Within ~5 ms post-impact, the entire mouse has fallen and is hanging by its grasped tail. Please click here to view a larger version of this figure.
Figure 3: Measurement of cerebral blood flow by Diffuse Correlation Spectroscopy. (A) An optical sensor is gently manually held over the right hemisphere to measure blood flow in an anesthetized mouse. (B) Representative sensor placement on the right hemisphere. The outline of the sensor is represented as a dashed black rectangle, and the location of the source and detector fibers are in red and blue circles, respectively. The sensor is placed such that the short edge of the sensor lines up with the back of the eye and the long edge of the sensor aligns with the midline. Please click here to view a larger version of this figure.
Figure 4: Diffuse Correlation Spectroscopy data analysis. (A) Representative measured intensity autocorrelation curves, g2(τ), at baseline, pre-injury baseline (green) and 4 h after 5 closed head injuries spaced once daily (purple). The right shift in the curve from pre- to post-injury reflects a decrease in blood flow. (B) g(τ) data is acquired at 1 Hz for 5 s per hemisphere and repeated 3x/hemisphere. Each measured g
(τ) curve is fit to the semi-infinite solution to the correlation diffusion equation for a cerebral blood flow index (CBFi). (C) CBFi values across all frames and repetitions are averaged to obtain mean cerebral blood flow index for each hemisphere (denoted by horizontal black bar). Please click here to view a larger version of this figure.
Figure 5: Mouse brain microdissection. (A) After the brain is extracted from mouse, it is bisected along the dashed line. The left hemisphere is fixed for histology, and the right hemisphere is microdissected for pathology. (B) Sagittal view of the cortex of the right hemisphere. The right hemisphere is microdissected into corresponding color-coded regions. For analysis of freeze-sensitive proteins, it is optimal to sub-divide tissue sections prior to flash freezing. Please click here to view a larger version of this figure.
Figure 6: Illustration of Luminex procedure. (A) Add samples to fluorescently tagged beads. Beads are pre-coated with a specific capture antibody for each protein of interest. (B) Add biotinylated detection antibodies. Biotin-detection antibodies bind to the analytes of interest and form an antibody-antigen sandwich. (C) Add phycoerythrin (PE)-conjugated streptavidin (SAPE). SAPE binds to the biotinylated detection antibodies, completing the reaction. For phospho-proteins, an amplification buffer (only for phospho-protein assays) is added following the addition to SAPE to enhance the assay signal. (D) Luminex instrument (MAGPIX, 200, or FlexMap 3D) reads reaction on each fluorescently tagged bead via a combination of red/green illumination. Please click here to view a larger version of this figure.
Figure 7: Illustration of sample dilution curve to identify the linear range. Protein concentration of serially diluted samples versus fluorescent intensity measured from the Luminex assay. The linear range is defined as the protein concentration range for which the relationship between the protein concentration and fluorescent intensity is linear (arrow). In some analytes, increasing the protein concentration beyond a certain limit can decrease antibody binding such that the dilution curve becomes non-linear or inverted (Hook effect). Please click here to view a larger version of this figure.
Figure 8: Representative Partial Least Squares Regression (PLSR) Analysis. (A) Panel cytokine protein expression (left columns) together with Iba1 expression (righthand column) in 3xCHI mice (n=8, z-scored). (B) PLSR assigns weights (loadings) to measured cytokines for each latent variable. Weights are applied to measured data to compute scores for each sample on each latent variable. (C) PLSR of 3xCHI samples against Iba1 identified a weighted profile of cytokines, LV1, which distinguished samples by Iba1. Cytokines with negative weights were upregulated in samples with low Iba1 while cytokines with positive weights were upregulated in samples with high Iba1 (mean ± SD using a LOOCV). (D) Linear regression of LV1 scores for each sample against Iba1. R2PLS measures the goodness of fit between Iba1 and LV1. (E) Individual regressions of Iba1 against each of the cytokines with the greatest weights in LV1 in C. Please click here to view a larger version of this figure.
Herein we detail methods for assessment of the hemodynamic and neuroinflammatory response to repetitive mild traumatic brain injury. Further, we have shown how to integrate these data as part of a multivariate systems analysis using partial least squares regression. In the text below we will discuss some of the key steps and limitations associated with the protocol as well as the advantages/disadvantages of the methods over existing methods.
Weight-drop model of mild traumatic brain injury. This method of traumatic brain injury induction is advantageous in that it features blunt-impact followed by rapid anterior-posterior rotational acceleration commonly seen in sports-related head injuries10,19. Certainly, the lissencephalic mouse brain does not fully recapitulate complexity of the gyrocephalic human brain; nevertheless, this model does induce many of the same clinical and behavioral sequelae of human mTBI, including sustained deficits in spatial learning and memory with repeated injuries. Additionally, while the impact is mild in nature (no structural/neuronal damage, no blood brain barrier permeability, cognitive deficits emerging only after multiple hits, etc.19), it does induce significant loss of consciousness, in contrast to humans where loss of consciousness is less common. This increased loss of consciousness may be due to an interaction with the anesthesia given immediately prior to the impact, although the exact cause is not well understood. Finally, we note that aligning the guide tube such that the bolt impacts between the coronal and lamdoid sutures is critical. We have observed that impacts that are more posterior can cause significant motor deficits that require euthanasia.
Assessment of cerebral blood flow with Diffuse Correlation Spectroscopy. Non-invasive, longitudinal measurements of cerebral blood flow (CBF) with traditional modalities used in human/large animal studies, such as perfusion magnetic resonance imaging or transcranial Doppler ultrasound, are challenging in mice for various reasons, including small brain size and total blood volume29. Diffuse correlation spectroscopy is well-suited in mice and offers the added advantages of being noninvasive and relatively inexpensive compared to other modalities20,30. Because DCS is sensitive to motion artifacts, mice need to be briefly anesthetized or restrained31 for assessment. We typically use isoflurane anesthesia due to its fast induction and recovery; however, isoflurane is a cerebral vasodilator, and blood flow estimates under isoflurane should be interpreted with caution. Decreases in blood flow seen post-injury compared with sham-injured animals could be confounded by a failure of the injured cerebral vasculature to vasodilate in response to isoflurane. Finally, we have previously demonstrated excellent intra-user repeatability of blood flow measurements with DCS in mice but only fair intra-user repeatability21. For this reason, we recommend that the same operator acquire DCS measurements for experiments that require longitudinal assessment of cerebral blood flow.
Multiplexed quantification of cytokines and phospho-proteins using Luminex assays. A key challenge with any ELISA is the Hook effect, whereby increased protein concentration can reduce antibody affinity for the target protein, thus leading to decreased assay signal in response to increased protein32 (Figure 7). This effect can be exacerbated when analyzing whole tissues, wherein bulk proteins can similarly interfere. Thus, the first step in utilizing Luminex assays is to determine if there is a range of protein concentrations loaded for which the assay read out linearly varies with the amount of protein loaded. Analytes that do not have such a linear range (Figure 7) should be excluded from analysis. We also note that cytokine levels in the brain are typically very low and appear near the lower detection limit assessed via standard curves supplied with the Luminex kit. For this reason, it is essential to conduct linear ranging analysis to determine if the instrument readout truly reflects the amount of sample loaded. For key proteins of interest, this linear range analysis can be complemented with a spike recovery assay wherein recombinant protein is spiked into a sample, and linearity in the instrument reading is evaluated33.
Because neuroinflammation is regulated by diverse intracellular phospho-proteins and extracellular cytokines, it is critical to simultaneously measure a wide array of these proteins in order to understand the brain’s neural immune response to mTBI. Luminex multiplexed immunoassays enable simultaneous quantification of dozens of cytokines and phospho-proteins from a single sample, providing a holistic view of the tissue immune response post-injury. Although these analyses provide a broad view of cytokines/chemokines as well as phospho-proteins, the assay quantifies total amount of each protein from a tissue homogenate. Thus, it does not yield cell-type specific data. Cell-type specificity can be determined by follow-up immunohistochemistry to identify localization of top proteins of interest with markers for cell types (e.g., neurons, microglia, astrocytes, etc.)11.
Partial least squares regression analysis for data integration. The tissue response to mTBI is multifactorial, consisting of physiological changes in blood flow together with changes in the phagocyte/microglial activation marker Iba1, diverse cytokines, and phospho-proteins, among others11. Because of the multiplexed nature of the data collected, a systematic method is needed to account for the multidimensionality of the relationships between the various predictor variables. PLSR provides a suitable solution to this problem by identifying LVs that maximally identify the co-variance between the predictor variables and the outcome variable (e.g., Iba1 in Figure 8). Importantly, those proteins found to strongly correlate with the predictor variable (i.e., those with high loadings on LV1) often correlate in the univariate regression analysis as well (Figure 8E). Because PLSR is often used to fit a large number of predictor variables to a smaller number of samples as in Figure 8, it is critical to gain an understanding of the sensitivity of the weights on LV1 to individual samples. For a small number of samples (<10) LOOCV is useful for assessing the sensitivity of the weights (indicated via SD error bars in Figure 8C). For a larger number of samples, it will be important to assess sensitivity by leaving multiple samples out at a time using a Monte Carlo sub-sampling approach34. We refer the reader to Multi and Megavariate Data Analysis24 for an in-depth discussion of PLSR approaches and uses. Finally, we note that a key limitation of this type of analysis is that it is purely correlative. PLSR does not prove a mechanistic relationship between predictor variables and the outcome variable. We view PLSR as a valuable hypothesis generating approach that is used to suggest tractable targets to modulate in future experiments that establish causal relationships.
The authors have nothing to disclose.
This project was supported by the National Institutes of Health R21 NS104801 (EMB) and R01 NS115994 (LBW/EB) and Children’s Healthcare of Atlanta Junior Faculty Focused Award (EMB). This work was also supported by the U.S. Department of Defense through the Congressionally Directed Medical Research Programs under Award No. W81XWH-18-1-0669 (LBW/EMB). Opinions, interpretations, conclusions and recommendations are those of the author and are not necessarily endorsed by the Department of Defense. This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. 1937971. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
Adjustable pipettes | any adjustable pipette | ||
Aluminum foil | VWR | 89107-726 | |
Bio-Plex cell lysis kit | C Bio-Rad | 171304012 | |
BRAND BRANDplates pureGrade Microplates, Nonsterile | BrandTech | 781602 96 | |
Complete mini protease inhibitor tablet | Sigma-Aldrich | 11836153001 | |
Depilatory cream | Amazon | Nair | |
DiH2O | VWR | VWRL0200-1000 | |
Handheld magnetic separator block for 96 well flat bottom plates | Millipore Sigma Catalogue | 40-285 | |
Hardware Autocorrelator Board | www.correlator.com | Flex05-8ch | |
Isoflurane 250 mL | MED-VET INTERNATIONAL | RXISO-250 | |
Kimwipe (11.2 x 21.3 cm) | VWR | 21905-026 | |
Laboratory vortex mixer | VWR | 10153-838 | |
LabView | National Instruments | LabVIEW | |
Luminex 200, HTS, FLEXMAP 3D, or MAGPIX with xPONENT software | Luminex Corporation | ||
Luminex Drive Fluid | Luminex | MPXDF-4PK | |
Luminex sheath fluid | EMD Millipore | SHEATHFLUID | |
MILLIPLEX MAP Mouse Cytokine/Chemokine Magnetic Bead Panel – Premixed 32 Plex – Immunology Multiplex Assay | Millipore Sigma | MCYTMAG-70K-PX32 | |
MILLIPLEX MAPK/SAPK Signaling 10-Plex Kit-Cell Signaling Multiplex Assay | Millipore Sigma | 48-660MAG | |
Mini LabRoller rotator | VWR | 10136-084 | |
Phenylmethylsulfonyl fluoride | Sigma-Aldrich | P7626-1G | |
Phosphate-buffered Saline (PBS) | VWR | 97064-158 | |
Plate Sealer | VWR | 82050-992 | |
Polypropylene microfuge tubes | VWR | 20901-547 | |
Mini LabRoller | Millipore Sigma | Z674591 | |
Reagent Reservoirs | VWR | 89094-668 | |
R Programming Language | |||
RStudio | www.rstudio.com | ||
Sonicator | |||
Titer plate shaker | VWR | 12620-926 | |
Tween20 | Sigma-Aldrich | P9416-50ML | |
1 m acrylic guide tube | McMaster-Carr | 49035K85 | |
4 photon counting avalanche photodiode | Perkin-Elmer | SPCM-AQ4C-IO | |
400 um multimode source fiber | Thorlabs Inc. | FT-400-EMT | |
54 g bolt | Ace Hardware | 0.95 cm basic body diameter, 2 cm head diameter, 10.2 cm length | |
780 nm single mode detector fiber | Thorlabs Inc. | 780HP | |
852 nm long-coherence length laser | TOPTICA Photonics | iBeam smart |