In vivo characterisation of endogenous cardiovascular extracellular vesicles in larval and adult zebrafish

Ischaemic cardiac injury results in a complex coordinated cellular response that enables effective damage control and recovery. Intercellular communication is integral to the coordination of these repair responses in complex tissue contexts. Extracellular vesicles (EVs) facilitate molecular transport across extracellular space, allowing local and systemic signaling during homeostasis and in disease. Extensive in vitro studies have described the biogenesis, cargos and functional roles of EV populations but the characterisation of endogenously produced EVs in vivo is still in its infancy. Here we use larval and adult transgenic zebrafish to characterize endogenous cardiovascular EVs. Using a membrane-tethered fluorophore reporter system, cell-type specific EVs can be tracked in vivo by high spatiotemporal resolution light sheet imaging and modified flow cytometry methods allows these EVs to be extracted and assessed ex vivo. Importantly, we show that ischaemic cardiac injury models dynamically alter EV production during repair.


Introduction
Extracellular vesicles (EVs) collectively describes all plasma membrane-bound vesicles produced and released by most cell types. An assortment of lipids, proteins and nucleic acids can be loaded into EVs either as luminal cargo or as a component of the EVs membrane providing signaling capacity for intercellular communication. EVs can be broadly split into three main classes: exosomes, formed from an endocytic pathway, microvesicles that are directly shed from the plasma membrane, and apoptotic bodies, which are shed from cells undergoing apoptosis (Caruso and Poon, 2018;van Niel et al., 2018). EVs can be trafficked both locally and systemically and have been isolated from a wide range of biological fluids, including blood (Crawford, 1971;Emanueli et al., 2016) and pericardial fluid (Beltrami et al., 2017;Kuosmanen et al., 2015). Various surface glycans, lipids and proteins work in combination to target EVs to regions of extracellular matrix (ECM) (Rilla et al., 2019) and recipient cells, and these interactions are widely recognised to play integral roles in various communicatory pathways (Colombo et al., 2014). EVs are most often considered within pathophysiological contexts, having been implicated in the progression of many diseases, including cancer and cardiovascular disease Tkach and Thery, 2016), however EVs also play roles in homeostasis (Ridger et al., 2017;Yanez-Mo et al., 2015). Cardiovascular disease remains the single biggest killer in the western world.
Myocardial infarction (MI), often as a result of coronary heart disease, drives ischaemia and ultimately cardiomyocyte cell death in affected regions of the heart, with the ensuing cellular events, including the inflammatory response and revascularisation, serving to repair the injured region and restore tissue function, but ultimately resulting in the formation of irreversible scarring and reduced function of the injured region (Li and Izpisua Belmonte, 2016). These repair processes require the concerted effort of multiple different cell types within the heart, implying the need for a complex communication system. In humans the dead myocardium is replaced primarily by fibrotic scar tissue limiting the functional capacity of the heart and leaving the patient vulnerable to future complications (Cahill et al., 2017;Li and Izpisua Belmonte, 2016;Talman and Ruskoaho, 2016). In regenerative species such as the zebrafish, however, the heart follows a similar initial repair process, but the deposited scar tissue is quickly resolved and the functional myocardium is fully replaced by cardiomyocyte dedifferentiation and re-entry into the mitotic cycle (Chablais et al., 2011;Gonzalez-Rosa et al., 2011;Poss et al., 2002;Schnabel et al., 2011). Zebrafish have emerged as a powerful model of human disease and have many advantages as a vertebrate model including fecundity, external development, genetic tractability and unrivalled cellular level in vivo imaging due to their transparency during embryonic and larval stages (Lieschke and Currie, 2007). Currently, little is known about the role of endogenous EVs in regenerative contexts despite the potential to identify pro-regenerative signals being exchanged between cell types.
The potential of EVs as biomarkers of disease and as novel therapeutic delivery vehicles has generated significant interest in recent years, including applications in tissue (predominantly bone) regeneration (Wiklander et al., 2019). However, the ability to reliably define the heterogeneous spectrum of EV subtypes, and further, to ascribe functional significance to that specific subtype in vivo, is still in its infancy (Beer and Wehman, 2017;Hyenne et al., 2017). The majority of EV characterization to date, by necessity, has been performed in vitro (van Niel et al., 2018). Strategies have been developed to obtain EV populations from culture medium, providing pure samples of EVs derived from known cell types, or from biological fluid samples, producing EVs from specific sites but from heterogenous origins, allowing the molecular components of these EVs to be determined and compared (Gyuris et al., 2019;Kalra et al., 2012;Thery et al., 2006;van Niel et al., 2018). Recent studies have developed novel ways to label exogenous EVs to investigate the biodistribution of these EV populations and are starting to dissect the functional changes elicited in recipient cells resulting from their addition to cell culture or in vivo systems (Beltrami et al., 2017;Hyenne et al., 2019;Lai et al., 2014;Wiklander et al., 2015). Whereas these studies using exogenous EVs are beginning to identify important roles in multiple different tissues and disease states, they fail to address the full complexity of endogenous in 5 vivo EV populations. A recent report has started to bridge this gap by investigating endogenously produced EVs in a novel zebrafish model (Verweij et al., 2019b). However, in vivo studies of endogenous cardiovascular EV function during disease (and potentially tissue regeneration) remain limited.
Here, we describe techniques to stably label endogenous cell-type specific EVs with fluorescent reporters in vivo in larval and adult zebrafish, allowing these vesicles to be tracked, extracted and validated. We demonstrate that global zebrafish EVs (either labelled through a ubiquitous promoter or by secreted Annexin-V binding) as well as endothelial-and cardiomyocyte-derived EVs can be observed in the peripheral circulation and the pericardial space, respectively. Using adapted flow cytometry techniques (including fluorescence activated vesicle sorting -FAVS (Cao et al., 2008)) and imaging flow cytometry (Lannigan and Erdbruegger, 2017)) we can analyse and purify these cell-type specific EVs from tissue samples and evaluate different EV populations ex vivo. Most importantly, we demonstrate that models of ischaemic injury, in larvae and adults, result in dynamic changes to the EV profile both systemically and specifically in the heart.

Stable labelling and imaging of endogenous EVs in vivo
EVs have been shown to be involved in multiple different developmental, homeostatic and disease situations including cancer and cardiovascular disease Tkach and Thery, 2016). Despite recent advances in labelling methods, the distribution and function of cell type specific EVs in vivo remains incompletely understood largely due to the technical challenges of studying these small vesicles. Using EV labelling methods similar to those previously described for exogenous vesicles (Lai et al., 2015), we have made use of stable transgenic zebrafish to label endogenous EVs. Using different promoters to drive the expression of prenylated fluorophores that are incorporated into the plasma membrane ubiquitously or in a cell type specific manner, we show that this label is incorporated into the EVs produced by the labelled cells. Initially, we used a transgenic line where prenylated 6 GFP is driven by a near-ubiquitous promoter, tethering the fluorophore to the inner leaflet of the plasma membrane and, consequently, incorporating GFP in the EVs from the majority of cells (Tg(Ola.Actb:Hsa.HRAS-EGFP) (referred to as Tg(actb2:HRAS-EGFP)); Figure 1A,B; (Cooper et al., 2005)). EVs labelled in this way were observed in the peripheral circulation and in the pericardial space of larval zebrafish at 3 days post fertilisation (dpf) (( Figure 1C Figure S1), including those relevant in cardiovascular disease (Jansen et al., 2017;Jansen et al., 2012). As with the actb2 line we observed Annexin-V labelled EVs in both the peripheral circulation and pericardial space of larval zebrafish ( Figure S1C,D).
Although these systems allowed us to successfully identify endogenous EVs in vivo, they do not allow us to determine the cellular origin of the EVs. To do this we made use of two more transgenic lines that express a prenylated fluorophore driven by cell specific promoters, whereby the fluorophore is incorporated into the EVs in a cell-type specific manner ( Figure 2). Analysis of larval zebrafish expressing endothelial (Tg(kdrl:mCherry-CAAX); EC-EVs) or cardiomyocyte (Tg(myl7:HRAS-mCherry); CM-EVs) specific promoters driving membrane tethered mCherry (Fujita et al., 2011;Yoruk et al., 2012) revealed the presence of cell-derived EVs in the peripheral circulation and pericardial space, respectively

Validation of endogenous EVs
To further characterise these endogenous EVs we sought to validate the identified populations further (Figure 3). We used a tissue dissociation protocol followed by differential centrifugation and filtration to separate a crude "large" EV fraction (centrifugation at 20,000 g) and a crude "small" EV fraction (centrifugation at 100,000 g) from entire zebrafish larvae  Figure 3G,H). Flow cytometry analyses suggest that 29 % ± 8.6 (SD) of total events are labelled with calcein, representing intact EVs ( Figure 3H).
Live imaging analysis cannot determine the true size of the EVs observed in the DA and pericardial space as these are likely to be below the resolution limit of light sheet microscopy (≥250 nm). To investigate the size of the extracted EVs further, we first analysed isolated fluorescently labelled EVs from Tg(actb2:HRAS-EGFP); (Tg(kdrl:mCherry-CAAX) double transgenic larvae by high resolution Total Internal Reflection (TIRF -approx. max resolution <100 nm) microscopy together with 60 nm synthetic beads revealing actb2(GFP)+ EVs of a similar size to the synthetic beads ( Figure 3I,J). Despite the fact these beads (and likely most EVs) are below the resolution limit, they can be visualised by their fluorescence, although their true size still cannot be accurately determined. TIRF imaging of a single sorted (FAVS) vesicle confirms triple labelling for calcein (intact vesicle), GFP (nearubiquitous marker) and mCherry (endothelial specific) ( Figure 3J). Finally, western blot analysis with an antibody against a known EV component demonstrated expression in isolated small and large EV fractions ( Figure 3K).

Ischaemic injury models result in changes to the EV profile of zebrafish larvae
We can observe and isolate EVs derived from specific cardiovascular cell types using our zebrafish model, but we wanted to determine if and how these vesicles are affected by models of cardiovascular disease and cardiac damage. Initially, we used a MicroPoint ablation laser to induce damage specifically to the surface of the ventricle in 3 dpf zebrafish larvae and assessed the number of EVs produced by flow cytometry (Figure 4A,B). We did not see changes to the overall number of calcein+ EVs at 6, 12, 18 or 24 hours post heart injury (hphi), perhaps due to the minor nature of the induced injury and/or limits to the sensitivity of flow cytometry techniques to detect subtle changes to EV populations. Next, we Supplementary videos 6,7). This suggests that both a substantial flank injury and a global ischaemic/hypoxic environment can induce increased EC-EV release into the peripheral circulation. Collectively, these data suggest dynamic EV responses following models of tissue injury in larval zebrafish, however, we also wanted to assess endogenous cardiovascular EVs from fully differentiated, adult tissues and to determine if a model of MI in adult zebrafish hearts resulted in changes to cardiovascular EVs.

Characterisation of EVs from adult cardiac tissue reveals dynamic changes following a model of MI
To validate EVs from adult cardiac tissue firstly we extracted EVs from isolated ventricles of unwounded adult fish and analysed these ex vivo using similar techniques described for larval samples ( Figure 5A). To further validate these adult cardiac EVs, we isolated total EVs from ventricles and visualised them by transmission electron microscopy (TEM; Figure   5B,C). This analysis revealed EVs with a characteristic cup-like structure in sizes ranging from ~50-300 nm ( Figure 5C,D), supporting our DLS data on size distribution of EVs in larvae. Flow cytometry analysis revealed populations of intact, cell-type specific EVs derived from adult ventricles that could be obtained in much higher numbers than from entire larval fish ( Figure 5E-I). To observe cell type specific EVs in vivo, we performed live imaging of superficial vessels of adult Tg(kdrl:mCherry-CAAX) fish revealing EC-EVs present in the peripheral circulation ( Figure 5J and supplementary video 8).
To assess the EV response to a model of MI, we performed cryoinjury on the hearts of adult zebrafish and extracted EVs from isolated ventricles of unwounded and 24 hpi fish and analysed them via flow cytometry and DLS ( Figure 6). Flow cytometry did not reveal significant changes to the overall number of calcein+ or intact CM-EVs in the heart following cardiac injury ( Figure 6A,B), however there were significantly fewer EC-EVs as a proportion of calcein+ EVs ( Figure 6C) suggesting an increase in EVs derived from a cell type not assessed here e.g. interstitial fibroblasts or inflammatory cells. Interestingly, DLS revealed a significant shift in the size distribution of total EVs at 24 hpi when compared to unwounded hearts ( Figure 6D) suggesting a dynamic EV response following cardiac injury. Here we describe techniques to stably label endogenous cardiovascular EVs.

Discussion
Endothelial cell derived EVs were observed in the peripheral circulation of larval and adult zebrafish, moving rapidly with the blood flow, but also exhibiting reduced speed when observed close to the vessel wall, as has been previously described (Hyenne et al., 2019;Verweij et al., 2019b). Cardiomyocyte-derived EVs were observed in the pericardial fluid of zebrafish larvae, where they were influenced by the movement of the heart. Static CM-EVs were also observed, potentially interacting with cells or ECM of the pericardial wall. We further analysed these endogenous EV populations ex vivo and show that large numbers of endogenous EVs can be isolated from zebrafish tissues using a modified FACS procedure.
We have also demonstrated that models of ischaemic injury result in dynamic changes to different EV populations.
The labelling strategy described here allows the identification of the cellular origin of EVs and to visualise these in vivo but the proportion of EVs that incorporate fluorophore labelling during biogenesis is incompletely understood, potentially limiting the number of EVs produced by each cell type that we can successfully identify (Shen et al., 2011). Indeed, flow cytometry assessments suggest that approximately 30-65 % of events are labelled with calcein (e.g. intact vesicles that contain esterases) and that approximately 30 % of these are labelled with actb2(GFP)+ (thought to be a near-ubiquitous promoter), suggesting relatively restricted fluorophore and esterase incorporation and/or limited preservation of intact EVs during isolation and/or sorting. Nevertheless, shared fluorescence between the selected cell types and EVs observed within extracellular space confirm the cellular origin of these vesicles and sufficient numbers of intact vesicles can be obtained for further studies, even when selecting only EVs that are both calcein+ and membrane-fluorophore labelled.
Larval in vivo imaging studies revealed static CM-EVs on the pericardial wall.
Analysis of the pericardium of Tg(actb2:HRAS-EGFP); Tg(myl7:HRAS-mCherry) larvae failed to show any direct interaction between CM-EVs and the cells lining the pericardial wall, potentially suggesting these EVs are interacting with an ECM component of the pericardium (Rilla et al., 2019). The larval zebrafish model described here is well placed to allow the study of these EV-ECM and EV-recipient cell interactions further. Additionally, future studies in larvae and/or adults would allow the functional significance of this interaction to be studied during repair and regeneration. Interestingly, we did not observe CM-EVs in the peripheral circulation or EC-EVs in the pericardial space of larvae, even following injury to the heart, suggesting localised control of EV release or rapid local uptake.
We have also demonstrated that we can visualise and extract EVs from adult zebrafish. EC-EVs could be observed moving rapidly with the blood flow in superficial vessels of adult fish and TEM reveals a characteristic cup-shaped appearance of extracted vesicles from isolated adult hearts. TEM measurements and DLS analysis reveal similar size distributions for total EVs extracted from adult hearts and from whole larvae, and are in line with previous reports of EVs from zebrafish, mice and humans (Akbar et al., 2017;Emanueli et al., 2016;Hyenne et al., 2019;Lawrie et al., 2009;Sahoo et al., 2011).
Most importantly, we have demonstrated that models of cardiovascular injury can induce changes to EV number and size, suggesting a dynamic intercellular communication response that involves EVs, supporting previous in vitro models, clinical findings and studies in non-regenerative rodent models of MI (Akbar et al., 2017;Beltrami et al., 2017;Boulanger et al., 2017;Emanueli et al., 2016;Gupta and Knowlton, 2007;Loyer et al., 2018;Malik et al., 2013). Similar numbers of overall EVs and CM-EVs following cardiac injury but a reduction in EC-EVs, as a proportion of total (calcein+) EVs, suggest an increased proportion of EVs deriving from an, as yet underdetermined, cell-type. Indeed, CM-EVs and EC-EVs represent approximately 30 % of total calcein+ EVs in the adult heart, suggesting a significant contribution from other cell types. Interstitial cardiac fibroblasts, and/or inflammatory cells are potential candidates and future studies will determine the contribution of these other cell types to the EV composition of the repairing and regenerating heart. Interestingly, we observed an increase in EC-EVs following larval injury models and a reduction in EC-EVs following an adult MI model, suggesting dynamic changes to EVs that may be highly tissue-and time-point dependent. Further studies will be required to determine if the reduction in EC-EVs results from reduced numbers of ECs in the injured heart at 24 hpi or from dynamic changes to EC-EV production following injury. Our data further suggest that there is a significant shift to smaller EVs at 24 hpi potentially indicating increased exosome release. This is similar to what has been described for clinical plasma 13 samples in patients undergoing cardiac surgery (Emanueli et al., 2016) and may represent an active shift in EV biogenesis in response to MI. Additionally, further studies to define the cargo and composition of these post-ischaemic EVs could reveal their roles in tissue repair and regeneration and identify potential therapeutic targets.
In summary, we have shown that endogenously produced cell-type specific EVs can be stably labelled with fluorophores, revealing their cellular origin and allowing them to be observed and tracked in vivo. Large numbers of cell-type specific EVs can be obtained from adult hearts allowing future evaluations of cargo during homeostasis and in models of disease. Finally, models of ischaemic injury produce dynamic changes to EV size and numbers produced by different cell types demonstrating the role of these small vesicles in these processes.
A single needle-stick injury into the dorsal epaxial musculature anterior to the cloaca was performed at 75°, using a 30-gauge needle. For hypoxia experiments, 3 dpf larval zebrafish in Danieau's medium were incubated at 28°C in a hypoxia workstation (InvivO2 300, Ruskinn), deoxygenised by positive infusion of 5% CO2/95 % N2 gas mixture to maintain 5% oxygen (Santhakumar et al., 2012) for 18 hours. Cardiac injuries on adult zebrafish were carried out as described previously (Gonzalez-Rosa and Mercader, 2012). Briefly, fish were anaesthetised in 0.13% MS-222 (Sigma; A5040) and placed ventral side up in a pre-cut sponge soaked in aquarium water containing anaesthetic. A 4 mm incision was made through the skin and the pericardial sac directly above the heart to expose the ventricle. The ventricle was dried using a sterile cotton swab and a liquid nitrogen cooled probe applied to the exposed ventricle for 30 seconds. Adult fish used were aged 4 -18 months and fish were randomly assigned to control or test experiments using a random number generator. All lines are maintained according to standard procedures and all animal work is carried out in accordance with UK Home Office and local University of Bristol regulations.

Imaging
3/4 dpf and adult zebrafish were anaesthetised in 0.01 mg.ml -1 MS-222 and mounted in 1% low-gelling agarose (Sigma). Live imaging was performed on a Zeiss Z.1 light sheet system with a 40x W Plan Apochromat objective. Additionally, some imaging was performed on a Leica TCS SP8 AOBS confocal laser scanning microscope with a 25x/0,95 W HC FLUOTAR objective. The resonant scanner option was used for high speed imaging (frame intervals of 0.02-0.04 seconds) of EVs in the peripheral circulation and the pericardial space. The conventional scanner was used for imaging of fixed samples. Images were processed using Fiji (Schindelin et al., 2012) and Imaris. Deconvolution was carried out using Huygens Professional version 16.10 (Scientific Volume Imaging, The Netherlands) using the CMLE algorithm. To achieve optimal results images were acquired at a sampling density that satisfied Nyquist-Shannon sampling theorem. Deconvolved images are noted in the figure legends. For manual counts of EVs, all analysis was blinded and positive events were counted from 1-minute videos of the DA above the cloaca (region highlighted in Figure 2C).

Cell Dissociation and EV Isolation
Adult or 4-5 dpf larval zebrafish were treated with an anaesthetic overdose before adult hearts were dissected (1-3 hearts pooled per experiment) or whole larvae (16-20 pooled per experiment) were placed into a digestion buffer consisting of perfusion buffer (PBS plus 10 mM HEPES, 30 mM Taurine and 5.5 mM Glucose) plus 0.25% Trypsin, 12.5 µM CaCl2 and 5 mg/ml Collagenase II (Worthington Biochemical Corp; LS004176). A stopping buffer consisting of perfusion buffer plus 10% (vol/vol) FBS and 12.5 µM CaCl2 was added to stop the digestion. EVs were isolated from the single cell suspension by differential centrifugation at 300 g (2 x 10 minutes), 1200 g (2 x 10 minutes), before being passed through a 0.8 µm sterile filter and a final centrifugation step of either 20,000 g (30 minutes) or 100,000 g (70 minutes) was used to crudely pellet large EVs and/or small EVs respectively, which were then resuspended in 300 µl sterile filtered PBS. Samples were either used immediately or snap frozen in liquid nitrogen, stored at -80 °C and later thawed at 37 °C for 2 minutes before use.

Dynamic Light Scatter (DLS)
Hydrodynamic particle size of crude EV samples in PBS was measured by DLS. All measurements were obtained in triplicate using a Zetasizer Nano-ZS equipped with an internal Peltier temperature controller (Malvern Instruments, Malvern Hills, UK).

Flow Cytometry
From crude EV samples, intact EVs were labelled with calcein violet 450 AM (eBioscience; 65-0854-39) as previously described (Gray et al., 2015). Detergent treated negative controls were mixed with 0.05% Triton X-100 (Sigma: T8787). Flow cytometry analysis and FAVS was carried out using a BD Influx system, with optimisations made to allow for the reliable detection of fluorescently labelled EVs (van der Vlist et al., 2012). Briefly, a high power 200 mW 488 nm laser was used for optimal excitation of small particles and GFP, a 50 mW 405 nm laser was used to excite calcein 450, a 100 mW 552 nm laser was used to excite mCherry and the following bandpass filters were used to detect the respective emissions: 530/40 nm, 460/50 nm and 610/20 nm. A small-particle detector allowed for high sensitivity in detecting forward scatter (FCS) and a 0.45 threshold on a logarithmic scale was used. A 4 mm obscuration bar was used to optimally detect submicron particles. For flow cytometry of EVs, a 100 µm nozzle and 21 PSI was used. To increase sorting speed and throughput for FAVS, a 70 µm nozzle and 42.9 PSI was used following confirmation that the signal was not compromised by the reduced time travelling through the laser. EVs were sorted into 100 µl PBS. All flow cytometry experiments were performed at 4 °C.
Imaging flow cytometry analysis was carried out as previously described (Lannigan and Erdbruegger, 2017). Briefly, analysis was performed using a fully calibrated (ASSIST tool) ImageStream ®x Mk II (Amnis-Luminex, Seattle, USA). 405 nm and 488 nm excitation lasers, a 785 nm side scatter (SSC) laser, brightfield illumination and a six channel charge-coupled device camera with time delay integration were used. For maximum resolution and high sensitivity, fluidics were set at low speed and magnification was set at 60x (0.3 mm 2 / pixel).
Speed beads (1.5 µm diameter, carboxylated polystyrene microspheres) run continuously during data acquisition to maintain focus and image synchronicity. They form a discrete population based on their SSC and fluorescence properties and can therefore be easily identified ( Figure S2A) and eliminated from the sample post-acquisition ( Figure S2B,C). The ImageStream Data Exploration and Analysis Software (IDEAS ® 6.2 EMD Millipore, Seattle) was used to perform data analyses. A compensation matrix was applied to adjust for spectral overlap between channels.

Transmission Electron Microscopy (TEM)
Samples were prepared for TEM in alignment with established methods (Rikkert et al., 2019). Five microlitres of isolated EV sample was placed onto Formvar-carbon coated, glowdischarged, 300 mesh copper grids (prepared in house by Wolfson Bioimaging Facility staff) and left for 30 seconds to allow the contents to adhere to the carbon support, the buffer was then removed using blotting paper. 1% uranyl acetate was then placed onto the grid and left to negatively stain the sample for 30 seconds, before being removed using blotting paper.
The grids were left to air-dry before being visualised on a FEI Tecnai 12 120kV BioTwin Spirit TEM and images acquired using a FEI Eagle 4k x4k CCD camera.
Images were then acquired using a Leica AM TIRF multi-colour system attached to a Leica DMI 6000 inverted epifluorescence microscope.

Statistics
In all cases n numbers refer to biological replicates. All experiments were repeated at least twice. Raw data recording and analysis was conducted using GraphPad Prism6/7. Statistical significance was determined via nonparametric Mann-Whitney or Kruskal-Wallis/Dunn's multiple comparison tests or a randomised permutation test using total variation distance between two groups (details provided in figure legends). All statistical analysis was blinded.
In all cases error bars represent SD. For all data sets a Grubb's outlier test was performed and any significant outliers (alpha = 0.05) were removed. Precise p values are included in all graphs.   Statistical analysis in A-C: Two-tailed Mann-Whitney tests. Statistical analysis in D: a custom permutation test using total variation distance was used to test the null hypothesis that control and 24 hpi distributions were the same.