Endothelial Regeneration of Large Vessels is a Biphasic Process Driven by Local Cells with Distinct Proliferative Capacities

Abstract

The cellular and mechanistic bases underlying endothelial regeneration of adult large vessels have proven challenging to study. Using a reproducible in vivo aortic endothelial injury model, we characterized cellular dynamics underlying the regenerative process through a combination of multi-color lineage tracing, parabiosis, and single-cell transcriptomics. We found that regeneration is a biphasic process driven by distinct populations arising from differentiated endothelial cells. The majority of cells immediately adjacent to the injury site re-enter the cell cycle during the initial damage response, with a second phase driven by a highly proliferative subpopulation. Endothelial regeneration requires activation of stress response genes including Atf3, and aged aortas compromised in their reparative capacity express less Atf3. Deletion of Atf3 reduced endothelial proliferation and compromised the regeneration. These findings provide important insights into cellular dynamics and mechanisms that drive responses to large vessel injury.

eTOC Statement:

Quiescent endothelial cells are able to mount a robust mitotic response even in the presence of pulsatile and high velocity blood flow. McDonald et al., showed that regeneration of aortic inner lining involves a subset of cells with hidden proliferative capacity that undergo rapid and significant transcriptional changes.

Graphical Abstract

INTRODUCTION

Maintenance and repair of many adult tissues relies on the presence of a pool of progenitor cells frequently confined to a specific region, their niche, supported by a myriad of cell-cell and cell-matrix interactions (Wagers, 2012). In response to either injury, or as part of the continuous need for cellular renewal, these highly responsive tissue-specific progenitor cells replicate to provide new cells that differentiate into appropriate subtypes and promote tissue homeostasis (Li and Clevers, 2010). In contrast to this well-accepted view of renewal, the mechanisms associated with regeneration and repair of the inner lining of large blood vessels have remained enigmatic.

Studies that sought to understand renewal of the endothelial lining in vivo have been confounded by the difficulty to document cellular loss, as these cells are quickly removed by blood flow. Similarly, accurate quantification of proliferation is challenging due to its transitory, dynamic nature and the difficulty of imaging the thin lining in transversal sections. The few studies successful at capturing proliferation in large vessels used radiolabeled thymidine uptake and en face visualization. These studies suggested that the overall basal rate of endothelial cell replication is nearly negligible in adult healthy blood vessels (Caplan and Schwartz, 1973; Kunz et al., 1978). Interestingly, some of these studies revealed that the proliferative behavior of endothelial cells was uneven within the lining, meaning, the endothelium displayed discrete foci with high proliferative capacity (Schwartz and Benditt, 1976).

While extremely valuable, the studies above pre-dated recent advances in genetic techniques to label single cells for visualization of their progeny and much improved visualization methods. Since these advances, questions about endothelial cell division in the context of the vascular lining of large vessels have not been pursued. Instead, experts in vascular biology have focused on the mechanisms associated with the morphogenesis of the cardiovascular system and the subsequent angiogenic expansion of vessels during development and in specific pathological settings (Potente et al., 2011). In sharp contrast to these advances, the requirements for cellular growth and expansion after a vessel has been formed remain far less understood. Similarly, the operative regulatory pathways involved in vascular regeneration and repair of large vessels after injury have not been explored. This knowledge is essential, since damage to the endothelium of large vessels, often resulting from endovascular medical procedures or severe vessel pathology, is detrimental to organ viability.

In the context of regeneration of the endothelium, several critical questions are currently unanswered: How do endothelial cells in an adult vessel respond after injury to renew the endothelium? Can the intima be repaired simply through the generation of more endothelial cells? Is there a particular site/niche or do all cells have equivalent repair capacity to undergo division regardless of their anatomic location? Are there endothelial cells with differential proliferative capacity? Does blood flow play a role? Are there differences in the endothelial regeneration capacity upon aging? And what are the molecular mechanisms involved in endothelial regeneration? These questions provided the motivation to seek information about re-growth of endothelium in the setting of an adult and physiologically active vessel: the infra-renal aorta. We characterized the process of regeneration, identified the cellular source of the renewal using a stochastic four-color reporter system to retrospectively trace the origin of new endothelial cells, performed gene expression analysis and a series of mechanistic studies to clarify the key regulators of the process.

RESULTS

Mechanical denudation injury in the adult aorta induces a prompt regenerative response within the endothelial lining both upstream and downstream of blood flow

In order to study the regeneration of the endothelial lining in vivo, we developed an injury model in which a 0.8mm width segment of the aortic endothelium was removed by mechanical pressure imposed by a clamp (Figures 1A and Video S1). Inspection of the tunica intima showed complete loss of the endothelial lining where the clamp was placed, as per absence of endothelial cell junctional staining (Figure 1B). The regeneration of the endothelial layer was then followed over time. We noted steady converging advancement of intact endothelial lining both upstream and downstream of the injured region with respect to blood flow direction (Figure 1C). The 0.8mm width injury was completely regenerated within 3 days (Figure 1D) at a rate of approximately 10 µm/hr (Figure 1E).

Figure 1. Kinetics of Endothelial Regeneration.

(A) Schema of aortic injury and dissection procedure for en face imaging.

(B) Images demonstrating removal of endothelial cells by clamp injury, VE-Cadherin staining (red) and fibrin (green). Scale bar=50µm.

(C) Regeneration of the endothelial lining over time. Top, fibrin staining highlights endothelial injury (yellow dotted lines). Regenerated area shows residual dim fibrin (blue dotted lines). Second from top, endothelial cell coverage, VE-Cadherin (red). Third from top, merged images. Scale bar=800µm.

(D) Kinetics of lining regeneration. Denuded area (yellow area in C) decreases as the regenerated area marked by residual fibrin (blue area in C) increases. Each dot represents one animal/injury. n=5 animals/point (average +/− SD).

(E) Left, plot of regeneration, defined by regenerated area over original wound width. Straight lines indicate linear regressions; dotted lines indicate 95% confidence prediction values based on regressions. Right, rates of endothelialization along the vessel’s long axis. n=4 animals/point (average +/− SD).

(F) Polarization of endothelial cells during regeneration. VE-Cadherin (red) and Giantin (green) to mark Golgi. Areas shown in next panel are indicated by the white boxes. Blue dotted line: original clamped area (injury). Yellow dotted line: leading edges of regenerating endothelial lining. Scale bar=800µm.

(G) High magnification showing endothelial polarity as defined by Golgi complex (Giantin, green) in relation to the nucleus (DAPI staining, white). In regenerating regions, the Golgi complex aligns with the direction of regeneration rather than with flow. Upstream (panel c) and downstream (panel d) wound edges marked with white arrows indicate the location of the Golgi. Scale bar=20µm.

(H) DAPI (white), Notch1-ICD (green), and VE-Cadherin (red) at the regenerating front (yellow dotted line). Arrows indicate Notch1 protein at edge of the regenerative front during arterial endothelium regeneration. Scale bar=10µm.

See also Figure S1 and Video S1.

Surprisingly, we observed that blood flow had no effect on re-endothelialization as both the upstream and downstream fronts of endothelial cells advanced at similar rates. To determine endothelial cell polarity during regeneration, we evaluated the position of the Golgi complex in endothelial cells near the wound area using the marker Giantin (Figure 1F). Evaluation of non-injured regions either upstream or downstream of flow showed the Golgi aligned along the circumferential axis of the vessel (Figure 1Ga,b, Figure S1). However, within the regeneration fronts the Golgi was enlarged and oriented in the direction of the injury regardless of blood flow (Figure 1Gc,d, Figure S1). These data indicate that endothelial cell polarity is defined by the regenerative response rather than by flow forces.

To explore potential similarities between arterial lining regeneration and angiogenesis, we were prompted to examine Notch1 protein distribution in the cells at the leading edge. In sharp contrast to the process of angiogenesis, we found that endothelial cells in the leading edge exhibited high levels of Notch1 protein (Figure 1H).

The mechanisms that promote endothelial regeneration are likely to include a combination of cellular hypertrophy, proliferation and perhaps migration. In fact, we noted that in the recently regenerated intima, endothelial cells were on average 10.6µm longer than cells in the non-wounded area (Figure S1). Cross sectional evaluation indicated that cells in the regenerated region are also larger in volume, as shown by greater protrusion from the lumenal surface at 48 and 96 hours post-injury (Figure S1). These data suggest that cell elongation and hypertrophy are associated with the early events of the regenerative process. Importantly, endothelial cell contacts were retained and the leading front advanced as a collective group of cells.

Robust cellular proliferation kinetics drives regeneration both upstream and downstream of blood flow

Next, we investigated how regeneration occurred, specifically whether proliferation was involved (Figure 2A). The possibilities included both broad and sparse distribution of proliferating cells along the aorta, local pockets of proliferation in specific areas, such as downstream of flow, or in proliferative clusters (Figure 2B). Using the thymidine analog EdU to mark cells in S-phase along a time course of regeneration (Figure 2C), we found robust proliferation of endothelial cells confined to the first 4–7 columns of cells flanking the injury both upstream and downstream of blood flow. Importantly, proliferation was limited to the endothelial cells during the time intervals assayed as demonstrated by strict co- localization with the endothelial-specific transcription factor Erg (Figure 2C). While EdU incorporation was only allowed to occur for two hours, we found that a high percentage of cells were positive (approx. 37%, Fig 2D) by 48hrs post-injury indicating that many terminally differentiated cells were able to quickly re-enter the cell cycle. The wave of proliferation extended through the entirety of the regenerating area (as identified by the fibrin staining), but not beyond it (Figure 2E–F).

Figure 2. Distribution of Proliferative Cells in the Regenerated Zone.

(A) Timeline of experimental design and data points. EdU was injected 2hrs prior to sacrifice.

(B) Graphical representation of hypotheses.

(C) Colocalization of EdU+ (red) and Erg+ (green) nuclei. Aortae harvested 48hrs postinjury. White boxes (top) indicate areas of high magnification (bottom). Scale bars= 800µm (upper) and 20µm (lower).

(D) EdU+ nuclei fraction after a pulse of 2h, 48h post-injury. n=5 aortae evaluated 400µm² to the right and left of the wound (mean +/−SD).

(E) Images depicting localization of EdU+ nuclei (white) in the regeneration zone. Scale bar=800µm.

(F) Quantification and distribution of proliferating cells in the regenerating zone. Each row represents one aorta. Dotted line indicates geometric center of wound. Every positive cell was placed in the exact location in relation to the center of the wound. Negative values, position upstream of wound; positive values, downstream.

(G) Fraction of EdU+ cells in the regenerating front upstream and downstream of blood flow. Each dot represents one individual animal (average +/−SD).

(H) Plot of the estimate of endothelial cells generated prior to wound closure as determined by EdU (green area under curve) and number observed to be proceeding through the cell cycle following wound closure (blue area under the curve). Dotted lines represent upper and lower SD. Each dot is average of EdU+ cells/animal.

(I) Hyperdensity following wound closure is shown by Erg positivity (white). Shown is 5 days post-injury (dpi) and resolution of this hyperdensity at 2 months post-injury (mpi). Number (n) of nuclei in that area is shown on the bottom right. Scale bar=20µm

Proliferation (as per EdU incorporation) was noted in very few cells at 24hrs, but it became prominent at 48 and 72hrs post injury (Figure 2E–F). Interestingly the distribution of proliferative cells upstream and downstream of flow was equivalent (Figure 2G), in line with the average endothelialization rate (Figure 1C–E), validating our previous observations that direction of blood flow does not affect proliferative capacity or regenerative function. Further, we observed that the number of endothelial cells in S- phase peaked once the lining was restored (Figure 2H), resulting in a final density far exceeding that of adjacent uninjured areas (Figure 2I). Importantly, this hyperdensity is not equally distributed within the repaired wound, indicating that it might be a reflection of groups of cells (clones) with higher proliferative capacity than others. Given that cell density eventually normalizes (by two months post-injury) (Figure 2I), there is an excess of endothelial cells generated during lining regeneration which are eventually lost, suggesting pruning and release of excess cells, or perhaps their remnants, into the circulation.

Cells in the regenerated endothelial lining originate from injury-adjacent endothelial cells

While our experiments clearly indicated that the endothelial lining could be fully regenerated, its origin required additional scrutiny. Considering that the contribution of bone-marrow-derived circulating endothelial cell progenitors had been discussed in the context of angiogenic expansion, we employed a combination of lineage tracing and parabiosis experiments to investigate the origin of the newly generated endothelial cells. For the first set of experiments, we bred an endothelial cell specific VE-Cadherin Cre- ERT2 line (Cdh5-PAC- Cre^ERT2) to a reporter mouse (lox-stop-lox tdTomato) to enable pulsed genetic labeling of cells of the endothelial lineage immediately prior to injury (Figure 3A). Following injury and regeneration, we assessed the regenerated lining for the presence of unlabeled cells. The emergence of such cells would indicate the contribution of a potential non-endothelial stem or progenitor cell type, while the presence of labeled cells would indicate that regenerated lining derived from pre-existing differentiated (Cdh5- expressing) endothelial cells (Figure 3B). We were able to precisely quantify the numbers of tdTomato+ and tdTomato- endothelial cells through computational processing of confocal imaging of labeled aortae (Figures 3C, Video S2). Because the labeling was inducible, it was essential to compare the frequency of excised cells that expressed the reporter in the uninjured area against that in the regenerated area (Figures 3D, E) within the same aorta. We found that the range of recombination in the uninjured area varied from 85 to >99.5% after tamoxifen injection and that the frequency of unlabeled cells was unchanged between regenerated and uninjured areas, leading to a best estimate of 0% and an upper boundary of 3% for the fraction of regenerated endothelial lining originating from a separate, non-endothelial stem or progenitor lineage (Figure 3F). Of particular note, in the mice where the labeling of the endothelium was >99.5% efficient, our quantification indicated that <0.5% of cells were unlabeled in the regenerated lining; labeled endothelial cells clearly gave rise to essentially all of the regenerated endothelial lining.

Figure 3. Cellular Origins of Regenerated Endothelium.

(A) Genetic system used to label endothelial cells immediately prior to regeneration.

(B) Illustration indicating the possible outcomes of the experiment.

(C) Detection of tdTomato positive and negative cells based on Erg co-staining (white). Penetration of Cre (as per tdTomato) can be variable, and thus, endothelial-dependent regenerated area should be “normalized” to percentage of Cre penetration in the non- regenerated area of the same animal. Scale bar=20 µm.

(D) Aortae of mice sacrificed at 1 day and 30 days (right) post-injury (dpi). Top right shows mouse with >99% labeling efficiency; bottom right shows 94% efficiency. Scale bar=800 µm.

(E) Quantification of tdTomato positive cells within the uninjured area and the regenerated area per individual mouse using the method demonstrated in Video S2 Figure S3. n=9 mice evaluated.

(F) Mean and 95% confidence interval boundaries on the contribution by non-endothelial cells, based on paired t-test quantifications. n=9 mice.

(G) Diagram depicting the parabiosis of GFP+ and GFP- mice to create blood chimerism.

(H) Illustration indicating the possible outcomes of the experiment.

(I) Flow cytometry of blood from GFP+ (left) and GFP- (right) members of parabiosed mice to assess chimerism by GFP expression. Data is representative of all pairs used.

(J) Images from GFP+ (left) and GFP- (right) members of parabiosed mice assessing regeneration of endothelium detected by Erg staining (top) and CD45+ staining to identify inflammatory cells (bottom). Non-endothelial GFP+ cells in the aortic media are indicated (white arrows) and those were also always CD45+. Scale bar=800 µm.

(K) Images of GFP+ and GFP- members of parabiosed mice after regeneration (white arrows). Green outlines track GFP+ cell locations for co-localization with other stains. Some instances require examination in 3D to resolve (insets, white arrows). Scaled bar=10 µm.

(L) Quantitation of Erg and CD45 staining for all GFP+ cells in the regenerated lining of all mice examined. Two parabiosed pairs were quantified and the number of cells evaluated for each animal is indicated.

See also Video S2.

Although the lineage tracing experiments demonstrated that the cell type of origin for the regenerated cells was endothelial, it did not address the possibility that these endothelial cells could be arriving from distant sites and delivered to the injury by blood circulation, as has been proposed and debated (Purhonen et al., 2008). To address this question, we used pairs of GFP+ and GFP- mice that were surgically parabiosed to share a chimeric circulation (Figure 3G–I). In these mice we made symmetrical aortic injuries and allowed the endothelial lining to regenerate. Therefore, the presence of a GFP+ endothelial cell in the regenerated lining of a GFP- aorta could be inferred to originate from the circulation and vice versa. The experiment allowed for all circulating cells, including any potential circulating endothelial cells, to gain access to both injured aortic regions. To facilitate interpretation of the experiment, we only used mice that showed nearly perfect blood chimerism (Figure 3I). While we observed GFP+ cells at the site of injury (Figures 3J–K) in GFP- mice and vice-versa, these cells were CD45+, localized to the arterial media (Figure 3K), and did not express endothelial marker Erg. Of note, some cases of closely apposed cells could only be disambiguated by taking advantage of the high resolution 3D confocal imaging data (Figure 3Ka,b). Evaluation of the regenerated area in parabiosed mouse pairs revealed that the regenerated endothelial lining contained no contributions from distant sites (Figure 3L).

Clonal tracing reveals differences in proliferative activity and provides evidence for transient-amplifying cell populations

Having established that the cells of the regenerated endothelial lining were created via proliferation of local endothelial cells, we asked whether there existed a hierarchy of cells with greater and lesser proliferative activity within the aortic endothelial lining. Alternatively, we thought that terminally differentiated endothelial cells might widely and non-selectively transition to a proliferative state to accomplish regeneration. Therefore, we used a multicolor fluorescent labeling allele (the “Rainbow” allele, Figure 4A) to evaluate the proliferative activity of single cells. Upon genetic recombination, the Rainbow allele stochastically yields permanent expression of three mutually exclusive fluorescent protein labels: Cerulean, mOrange, and mCherry. For visual contrast, we have represented these labels as cyan, green, and red respectively. These labels are inherited by daughter cells, thus allowing identification of clonally derived cells, resulting from the proliferation of a single genetically (and fluorescently) labeled ancestor cell.

Figure 4. Clonal Tracing Detects Cells with Distinct Proliferative Potential in the Endothelial Lining.

(A) Multicolor fluorescence (“Rainbow”) genetic labeling system used to label single endothelial cells. Mutually exclusive lox variants allow one of three fluorescent proteins to be expressed following tamoxifen-induced recombination.

(B) Graphical representation of possible outcomes.

(C) Images demonstrating clonal density by a highly diluted tamoxifen-induced labeling. The area of injury is visible as dimmed GFP fluorescence (white) (top panel). Representative clones in the uninjured (bottom left in “a”) and regenerated (bottom right in “b”) area of endothelial lining are shown at high magnification. Co-staining with Erg (white nuclei) allows for accurate clone size determination. Scale bar= 800µm (upper) and 20µm (lower).

(D) Computational extraction of labeled cell locations using the method demonstrated in Video S3.

(E) Spatial clustering algorithm dbscan with a distance threshold parameter of 150 microns was used to identify clones.

(F) Quantification of labeling frequency, calculated as (Cerulean/mOrange/mCherry)+/ERG+ nuclei in uninjured areas. Error bars indicate 95% c.i. n=4mice.

(G) Clone size distribution and curve fits. Scatter plot depicts cumulative frequency (Y- axis) of clones which are composed of n cells or less, normalized to the average n of the aorta. Clones labeled by the three different fluorescent proteins were pooled and each clone given equal weight. The dashed line displays the best fit of a negative exponential distribution, expected to arise from neutral competition between proliferating cells. The solid line shows the summation of two fitted curves, in effect modeling the presence of two populations with differing proliferative activities. n=8 aortae.

(H) Breakdown of the solid line in g, showing the two negative exponential curves which compose it. Shaded areas indicate the area under each curve, corresponding to the proportion of cells derived from each population.

See also Figure S2, S3 and Video S3.

If cells undergo neutral proliferative competition to replenish a tissue wherein no subpopulation of cells has a proliferative advantage, the cumulative probability of detecting a clone made up of n daughter cells follows a negative exponential distribution (Klein and Simons, 2011). We therefore asked whether the cumulative frequency distribution of endothelial clones of size n followed such a distribution, or if we could detect the presence of a more proliferative subpopulation of cells as evidenced by the selective outgrowth of a small number of clones (Figure 4B). Using a titrated tamoxifen dosage to achieve clonal labeling density, we systematically imaged a cohort of regenerated aortic linings (Figures 4C, S2, and Video S3) and extracted the coordinates of all labeled cells within the regenerated lining, (Figure 4D). We then used the spatial clustering algorithm dbscan (Ester et al., 1996) with a distance threshold parameter of 150 microns to identify clones composed of neighboring labeled cells within the regenerated aortic lining (Figure 4E). Importantly, we showed that labeling frequencies are low enough at 1% or less to support accurate clonal identification (Figure 4F).

To determine whether the data suggested the presence of one or two underlying populations, we separately fit a single negative exponential distribution and a double negative exponential distribution to the data. A single, best-fit negative exponential distribution (and thus a single population) does not sufficiently account for the long tail of large clones (Figure 4G), corresponding to infrequent clones which undergo massive proliferative expansions (Figure S3). The observed clonal distribution is better explained by the combination of two exponential distributions (Figure 4G,H), corresponding to two populations, each with a different extent of proliferation (p<10–4, loglikelihood ratio test). We estimate 70% of the new cells come from the actively proliferating subpopulation: transient-amplifying cells. This curve-fitting result was robust across multiple clustering distance thresholds and is particularly noteworthy since the mOrange and mCherry labels have lower labeling frequencies, and therefore more accurate clonal resolution. Based on the expansion pattern of over 250 endothelial cell clones tracked across 8 mice, our data indicate that the regeneration of the aortic lining is accomplished by at least two cell populations which exhibit functionally differing proliferative activities in the native vessel context.

Changes in the endothelial transcriptome during regeneration are distinct from sprouting angiogenesis

Next, we investigated the molecular programs activated during regeneration of the aortic lining. We performed comparative RNA-seq analysis on samples isolated 3hrs and 48hrs after injury from both uninjured control and wounded arterial segments using a lumenal flush method (Figure 5A). Importantly, we achieved a high degree of enrichment for endothelial cell specific transcripts when compared to the transcript fraction derived from the tissue remaining after lumenal flush (Figure S4). In the 48hr group specifically, we noted a high enrichment for transcripts associated with cell cycle activity (Figure S4). These data supported our previous findings that proliferation is an important component of the regenerative program for the endothelium, as opposed to only cellular hypertrophy and perhaps migration. Principle component analysis showed a clear separation between homeostatic and regenerating endothelial linings (Figure 5B). Differential expression analysis contrasting homeostatic and regenerating segments revealed that the endothelial lining reacts rapidly to changes imposed by injury. In fact, the transcriptome of cells adjacent to the injured area completely segregated from the control endothelium after a period as short as 3hrs post-injury (Figure 5B). Alterations in the transcriptional profile of endothelial cells were more impressive by 48hrs, a time of active regenerative activity (Figure 5C and Table S1).

Figure 5. Molecular Signature of Endothelial Regeneration.

(A) Experimental timeline. Aortae were harvested 3 and 48 hours following arterial injury. Also illustrated are the experimental samples: control, regeneration at 3h and regeneration at 48h. Note that the control (thoracic region) was isolated from the same animal as the regeneration injury at 48h. Uninjured thoracic and abdominal aorta segments (“tissue”) were also collected for comparison, the sample is referred to as “tissue” because there is no flush, meaning the entire tissue was used in the RNA extraction. RNA “flush” collection method which shows enrichment of endothelial RNA is illustrated at the bottom where lysis buffer is injected through the lumen enriching for endothelial transcripts. Cells hypothesized to actively drive lining regeneration are highlighted in red.

(B) Principal component analysis of collected transcriptomes. Note segregation of the four experimental groups. Tissue (purple), endothelial-enriched control (cyan), endothelial- enriched 3h post-injury (green) and endothelial-enriched 48h post-injury (red).

(C) Transcriptional signature of regenerating aortic lining. Transcription factors with a statistically significant (FDR = .01) change in transcript level greater than 4-fold are shown. Each column represents one aorta/individual animal: Control 48h, n=9; Regenerated 3h, n=5; Regenerated 48h, n=9.

(D) Images of injured aortae stained en face for select proteins to validate the information obtained by RNAseq analysis (C). Left column shows low magnification of aortae at 48hrs post-injury, in some cases the regeneration had just completed, VE-Cadherin and Erg (both in red). Middle column: expression pattern in an area adjacent to the injury (adjacent). Right column: expression in the regenerated area. Scale bar=20µm.

(E) Mice with endothelial-specific inactivation of Myc and FoxM1 were subjected to injury. Wound closure rate and proliferation kinetics were compared to control litermates (Cre negative or not injected with tamoxifen). VE-Cadherin (red), Erg (green) and EdU (white). High magnification (bottom right) corresponds to the white box in the image. Scale bar=100µm.

(F) Quantification of wound area and proliferation kinetics of the experiment shown in (E) for Myc effects. Wound closure evaluated at 48hrs and 96hrs hpi. Each dot represents one animal. Controls (black) and Myc^ECKO (red) were injected two weeks prior to the injury. 48hrs hpi n=13 for controls (Cre+ tam injected n=3; Myc^{lox/lox Cre-} tam injected=6; and Myc^ECKO non-injected=6) and n=7 for Myc^ECKO. Graph shows mean +/− SD. Statistics= Mann-Whitney test *p=0.0063. For 96hpi Myc^ECKO non-injected n=5 and Myc^ECKO tam- injected n=3, ns = not significant. Graph (right) shows proliferative responses as per % of Erg positive EdU positive cells over total EdU positive cells (same mice as the ones described above for 48hpi). Mann Whitney test p=0.0004.

(G) Quantification of wound area and proliferation kinetics for FoxM1 effects. Wound closure is evaluated at 48hrs and 96hrs hpi. Each dot represents one animal. Controls (black) and FoxM1^ECKO (red) were injected two weeks prior to the injury. N=9 for controls (Cre+ tam injected n=3; FoxM1^{lox/lox Cre-} tam injected=2; and FoxM1^ECKO non-injected=4) and n=9 for FoxM1^ECKO. Graph shows mean +/− SD. Statistics= Mann-Whitney **p=0.0078. For 96hpi FoxM1^ECKO non-injected n=5 and FoxM1^ECKO tam-injected n=3, ns = not significant. Graph (right) shows proliferative responses as per % of Erg positive EdU positive cells over total EdU positive cells (same mice as the ones described above for 48hpi). Mann Whitney test p=0.077.

See also Figure S4, S5 and Table S1.

Assessing the transcription factors altered during the process of regeneration revealed the activation of a “reaction-to-injury” program that included Fos, Myc/Mdx and the Hippo pathway (Tead4) as early as 3hrs (Figure 5C). Interestingly, we also found trends that indicated changes in arterial identity (Figure S5). In particular, we found statistically significant reduction of Nrp1 and Efnb2 with concurrent elevation in some lymphatic/vein markers, such as Lyve1, Prox1, and Nr2f2. Of the transcripts for genes involved in junctional integrity and mechanotransduction, Cdh5 and Gja4 were perturbed, while Gja5 was unaffected. Migration guidance molecule Sema3g was increased, as was Ly6a (Sca1)

At 48hrs post-injury, we observed a significant increase in expression of transcription factors associated with proliferation including E2f8, Foxm1 and Myc (Figure 5C). While much can be learned by further dissecting the transcriptomic profile (Table S1), the central objective of this analysis was to identify a molecular signature associated with the regenerative process. We proceeded to validate select candidate genes identified in the RNA-seq analysis by immunofluorescent staining for protein detection in the adult regenerating aorta (Figure 5D). FosL2, a member of the FOS/Jun family of proteins was found to be constitutively present at junctional borders in the quiescent, uninjured aorta. Interestingly, the transcription factor relocated to the cytosol and, to a lesser extent, the nucleus in the regenerating areas. Expression of Myc was perhaps the most impressive, nuclear localization of Myc was clearly observed in the 4–7 columns of endothelial cells flanking either regenerative front but was absent from the endothelium beyond that point. FoxM1, found to be transcriptionally increased by 48hrs, was clearly confined to areas of the regenerating endothelium and present at higher levels in the nucleus relative to the cytosol. Despite not detecting transcriptional changes in FoxO1, we found changes in localization. The protein was modestly increased in the cytosol, but largely excluded from the nucleus of endothelial cells participating in regeneration. Sca-1, a marker expressed by adult stem cells was clearly increased in the regenerating cells (Figure 5D).

To functionally test the relevance of these findings, we performed injury experiments in mice that lacked Myc and FoxM1 in the endothelium and assessed the progression of wound closure and proliferation kinetics by EdU incorporation (Figures 5E–G). The findings were gratifying as they demonstrated that absence of these genes affects would closure at 48hrs post-injury. Myc, but not FoxM1, also impaired proliferation as determined by EdU incorporation. Importantly, these findings suggested only a delay, as the wound of both model mice closed by 96hrs post injury (1–2 days after controls).

While the previous findings demonstrated the involvement of Myc and FoxM1 in endothelial regeneration, they came short of identifying a broader master regulator of the process. We thus, considered to apply single cell-RNAseq and inquire whether there were hierarchical differences within the aortic lining. Unfortunately, the initial comparisons between non-regenerating and regenerating cells simply reproduced the total RNAseq data. We then postulated that if indeed there is a hierarchical cellular difference in the endothelial lining this should be noted in an unwounded population and modified/reduced upon aging. To test if aging affected endothelial regeneration, we performed injury assays in young (8wks) and old (18mo) mice. Indeed, aged aortae showed a significant impairment in regeneration (Figure 6A). Using single cell RNAseq to evaluate uninjured young and old endothelium showed important transcriptional differences (Figure 6B, Figure S7). More noticeably, the list of transcription factors identified by RNAseq (3hpi) (Figure 5 C) revealed that Atf3 was far more abundant in the young than in the old endothelial lining (Figures 6C, D). In fact, staining for Atf3 in unwounded endothelium supports these results (Figure 6E). Furthermore, upon injury there is a higher frequency of Atf3 positive cells in young versus old aortae (Figure 6E). Next, we queried the single cell data for the transcriptional profile of Atf3 positive cells and compared it to Atf3 negative cells from the same aorta (Figure 6F). This highlighted a cohort of early stress cell response genes involved with regulation of proliferation such as Fox, Jun, Egr1, as well as Klf4 and Klf2. Interestingly, an unsupervised evaluation of upregulated genes for young versus old endothelium (Figures 6G, H) identified most of the genes found in the Atf3 positive population from the young endothelium (Figure 6F). Furthermore, a string analysis revealed functional and structural relationships between Atf3 and many of those genes identified in the aforementioned analysis (Figure 6I).

Figure 6. Cellular and Molecular Characterization of Endothelial Regeneration in Young and Aged Mice.

(A) Comparison of endothelial regeneration in young (8 weeks) and aged mice (18 months), VE-Cadherin(red), fibrin(green). Wound at 3hrs (upper) and 96hrs post-injury (lower). Bottom graph demonstrates the kinetics of wound closure in young (green dots) and aged mice (red dots). N=7 mice/time point. Mean +/−SD. Unpaired t-tests with Holm- Sidak correction for multiple comparisons (0h p=0.7509; 48h, 96h &120h ****p<0.0001). Bar=100µm.

(B) Disney plots of single cell RNAseq of the uninjured aortic endothelium of mice at 8 weeks and 18months of age. The endothelial cell cluster was revealed by expression of Pecam1 and Cdh5 as shown in the lower graphs.

(C) Transcription factors increased at 3hrs post injury in Figure 5 were evaluated against the uninjured aorta using the single cell data set. Atf3 was significantly increased in the young and nearly absent from the aged cohort, as shown in the heat map.

(D) Shown is the subpopulation of Atf3 positive endothelial cells in 5,000 endothelial cells from young (8wk) and old(18mo) mice.

(E) Immunofluorescence of Atf3 in unwounded and wounded aortae from young and aged mice. Bottom panel shows nuclear localization of Atf3 (green). Number of Atf3 positive cells near the wound is significantly reduced in old mice. Unwounded bars= 20µm. Wounded up and downstream bars=15µm, zoom bars=4 µm.

(F) Single cell transcriptome of cells that expressed Atf3 was compared to cells that do not expressed Atf3 from the same 8 week aorta. Note the genetic signature associated with Atf3 expression reflects a large number of genes associated with response to stress/injury.

(F) Comparison of the top upregulated and downregulated genes when the single cell transcriptomes of young versus aged mice are compared. Note the similarly between the unsupervised highly expressed genes in the young versus those genes associated with Atf3 expression in panel (E).

(G) Unsupervised heat map of the genes upregulated in young endothelium when compared to old endothelium. Note that all those cells are also found in the subset of cells from the young aorta panel that do not express Atf3 in (F).

(H) Unsupervised heat map of the genes downregulated in young endothelium when compared to old endothelium.

(I) String analysis reveals previously acknowledged relationships between ATF3 and several of the genes identified in (F) and (G).

See also Figure S6, S7 and Video S4.

The findings so far were consistent with the hierarchical presence of endothelial cell progenitors within the endothelial lining. In fact when freshly isolated, we noted that endothelial cells already show distinct proliferative capacities (Figure 7A, Movie S4). The large majority of the cells did not divide, only 24% of endothelial cells when placed in culture were able to divide 1 to 3 times in the span of 5 days (Figures 7B). The percentage of the dividing population was drastically reduced by 18 months when only 2% of the cells were observed dividing and even then, only once in the span of 5 days (Figure 7B). Interestingly, the fraction of endothelial cells using the Atf3 promoter, as per reporter assays was remarkably similar to the percentages of proliferating cells (Figures 7C,D,E). To test for causality rather than correlation, additional experiments explored the association of Atf3 promoter activity with cell division. We found that the large majority cells committed to proliferate (approximately 9% of all cells evaluated) expressed Aft3 (7% in average) when compared to (0.5%) of those not expressing Atf3 (Figure 7F, Movie S5). Meaning, close to 80% of the dividing cells expressed Atf3. Finally, we tested whether removal of Atf3 would affect endothelial regeneration. Mice that lack Atf3 showed a significant reduction in proliferative capacity (Figure 7G,H). Importantly, inactivation of Atf3 significantly impaired endothelial regeneration in young mice (8wks), in fact these mice resembled aged mice, as their wound remained open 96hrs post-injury while controls had closed by 48–72hrs post injury (Figure 7I).

Figure 7. Proliferation Kinetics and Atf3 Activity in Endothelial Cells.

(A) Proliferation kinetics of aortic endothelial cells isolated from young (8wks) mice. Endothelial cells labeled with tdTomato were plated onto unlabeled mouse endothelial monolayers and recorded using IncuCyte™. The large majority of individual cells did not divide, as exemplified by Clone 1; while a small cohort was able to divide multiple times, as exemplified by Clone 2. Scale bar=50µm. This type of evaluation was expanded to include additional endothelial cells isolated from multiple young and old mice and quantified in (B).

(B) Endothelial cell divisions from young (8wks) and old (18mo) mice were quantified over five days. The large majority of endothelial cells from the aorta of 8week mice do not divide (black), a few divided once (gray), less than 10% divide twice and approximately 12% divide at least three times. This contrasts the proliferative capacity of endothelial cells from older mice (18months). N=5 different aortas. Number clones evaluated: 353 (young) and 389 (old).

(C) Diagram of the lentiviral Atf3-eGFP reporter construct with nuclear localization signal (3XNSL). Aortic endothelial cells infected with the Atf3 reporter construct (green) and tdTomato were followed over time (indicated on the upper left corner). Cells were continuously monitored for 5 days in the IncuCyte. Dividing cells turned on reporter (green). Scale bar=75µm.

(D) Quantification of Atf3 reporter positive cells in young and old endothelium (from experiment shown in C). Each dot corresponds to one independent experiment in which 100 cells evaluated. A total of n=8 experiments were performed in each category. Data are presented as mean +/−SD Mann Whitney test p=0.0002.

(E) Low magnification view of eGFP reporter after cells from in young and old aortae were infected with the lentiviral vector. Arrows denote positive cells. Scale bar=300µm.

(F) Quantification of cells that show Atf3 reporter expression in young and old endothelium. Approximately 16% of cells were positive in young but only 2% of cells were positive in old endothelium. N=7 independent experiments. Data are presented as mean +/− SD. Mann-Whitney test p=0.0006.

(G) Endothelial regeneration in the absence of Atf3. Proliferation kinetics were evaluated using EdU at 48hrs post-injury. Representative images of aortae injury are shown stained with VE-Cadherin (red), Erg (green) and EdU (white). Bottom panel, EdU Images at higher magnification. Scale bar=100µm.

(H) Quantification of the proliferation kinetics in control and Atf3 knock out mice. N=8. Data are presented as mean +/−SD. Mann-Whitney test, p=0.0002.

(I) Quantification of wound closure 96hpi in control and Aft3 knock out mice. Control n=5, Atf3^KO n=4. Mann-Whitney test, p=0.007.

See also Video S5.

DISCUSSION

In this study, we utilized a combination of EdU labeling, parabiosis, cell tracing, transcriptional profiling and functional validation to identify the cellular and molecular events associated with the regeneration of the endothelium in vivo. We found that mitotically quiescent endothelial cells quickly sense and respond to a denudation injury by rapidly re-organizing their transcriptome and re-entering into the cell cycle. Importantly, the reaction is local, meaning about 4–7 columns of endothelial cells sense and carry out the complete regenerative process. A combination of EdU labeling and tracking of multi- fluorescently labeled clones indicate that the process is biphasic, including an initial phase when the large majority of cells in the regenerating front enter the cell cycle and a second phase when only a subset of cells continued to proliferate and conclude regeneration. The endothelial response includes a collective behavior of cells, that contrasts the prototypical tip-stalk cell characteristic of angiogenic growth. The regenerative response was significantly impaired in aging mice suggesting indicating a finite control of endothelial cell replication. Finally, we identified Atf3 as an important transcriptional regulator of the regenerative responses in arterial endothelium.

We provide several independent lines of evidence that the process of intimal repair and regeneration is prompted and fully executed irrespective of flow by endothelial cells located at both borders of the injury. The contribution and specific nature of endothelial cell progenitors has been a subject of heated debate (Purhonen et al., 2008) and therefore, we were committed to rigorously explore their potential contribution in this setting. Findings from adult endothelial lineage tracing using the VE-Cadherin promoter and double-injury parabiosis model support the conclusion that circulating endothelial cell progenitors or bone-marrow derived endothelial progenitors do not participate in events associated with endothelium regeneration of the aorta in mice. In all fairness, our two approaches do not rule out the contribution of for example, adventitial endothelial cells from the site of injury. Although given the fact that direct connections from the adventitia to the tunica intima are not present, this is an unlikely possibility. Circulating progenitor cells have been implicated in angiogenic expansion in a few settings (Asahara et al., 1997), it is possible that the high flow rate and inherent physical forces present in the aorta prohibit seeding and colonization of denuded vascular areas, although we find leukocytes in between and under the endothelial lining during regeneration (Shirali et al., 2018). More recently, elegant studies using side-cell population studies identified a cell surface marker for endothelial cells, CD157, that appears to identify a stem/progenitor population residing within the vascular wall of multiple vascular cells and organs (Wakabayashi et al., 2018). Our results in the aorta are consistent with that study and the presence of a hierarchy of endothelial cells that exhibits a unique transcriptional profile and enhanced ability to proliferate, although in the aorta, we did not find that CD157 labeled regenerating cells.

The process of endothelial lining regeneration in the aorta clearly involves collective cell behavior. Our findings indicate that endothelial cells retain their junctional complexes and remain connected throughout the entire repair process. Individual cell migration away from the front of regenerating endothelium was not observed.

An intriguing transcriptional signature of endothelial regeneration includes a subset of transcription factors that are associated with junctional complexes, in particular FosL2 and members of the Hippo pathway. Fox-like antigen 2 (FosL2 or Fra2) in particular has been poorly described with effects reported in bone development (Bozec et al., 2010) and photoperiodic regulation (Engel et al., 2005). Interestingly, FosL2 transgenic mice develop a severe proliferative vasculopathy of the lungs resembling pulmonary arterial hypertension (Eferl et al., 2008). These findings are consistent with our data in the aorta and expand the potential contribution of this transcription factor as a likely regulator of endothelial regeneration and growth in large vessels. A second transcriptional pathway with links to the cell membrane is the Hippo pathway. This pathway is a candidate as an initial trigger of endothelial regeneration since Yaz, a member of the Hippo pathway, has been shown to be localized at the cell surface and it is regulated by tension, in addition to controlling cell proliferation and organ size (Deng et al., 2015, Pan et al., 2016). We found that Tead4, the transcriptional partner of Yap, is increased shortly after injury, implicating the Hippo pathway in the early events associated with endothelial regeneration.

We also observed differences in a set of transcription factors known to regulate cellular proliferation during angiogenic sprouting and regeneration of lung vasculature, namely Myc, FoxM1, and FoxO1. The transcription factor Myc is tightly integrated with cell cycle progression. Myc accelerates the G1 phase of the cell cycle through promotion of Cyclin and CDK activity and suppression of CDK inhibitors p21 and p27 (Amati et al., 1998). During regeneration, we show upregulation of Cyclins and CDKs, and suppression by 48 hours of p21 transcription (Figure S4). We observed a rapid transition of endothelial cells from a quiescent differentiated state to rapid coordinated proliferation that coincided with changes in transcriptional markers of cell identity. This shift in identity is temporary, as it resolves 1month post injury (Shirali et al., 2018). Importantly, these changes in transcriptional profile occur as early as 3hrs after injury and include the reduction of arterial markers (Nrp1 and Efnb2) and the gain of venous and lymphatic markers (Nr2f2, Lyve1, Prox1, and Flt4). These discrete changes suggest the return to a more primitive stage that was perhaps more permissive for entry into the cell cycle. While this is speculative, it has been demonstrated in zebrafish that endothelial cells from veins are able to proliferate (and migrate) more than arterial endothelial cells (Xu et al., 2014). The study by Xu and colleagues demonstrated that the mechanisms of arteriole regeneration are highly dependent on the proliferative activity of veins. Therefore, it appears that the mechanisms of vascular regeneration might be broad and employ distinct cellular sources depending on the size, location and perhaps organ-type. Along these lines, we observed dramatic upregulation of the normally poorly expressed transcription factor FoxM1 as regeneration proceeded. This finding is notable because FoxM1 has been reported to regulate cell cycle progression and endothelial cell regeneration in the lung vasculature (Huang et al., 2016), and also because the related factor FoxO1 has been shown to direct Myc activity in the angiogenic retina (Wilhelm et al., 2016). Taken together, these data suggest a broad role for FoxM1 and Myc in controlling regeneration and phenotypic switching with FoxM1 potentially regulating the activity of Myc. Endothelial deletion of both Myc and FoxM1 delayed endothelial regeneration but alone they were unable to completely interrupt the process.

Clearly the main cellular process that drives regeneration in the injured aorta is a robust proliferative response coordinated by a cohort of transcriptional activators. Approximately a third of the cells located at the injury border were captured in S-phase by EdU incorporation 48 hours post injury, implying nearly synchronic activity and revealing proliferative plasticity. We found no indication that the ability of these cells to enter the cell cycle was confined to a particular geographical niche in the aorta, in fact proliferating cells were located immediately at the border of the injury site. These findings are consistent with the absence of a specific region (niche) or a special population of progenitor cells for the purposes of renewal and regeneration. Instead, the endothelium of large arteries relies on its own mitotic capacity. However there appears to be some hierarchy within this capacity. While the initial mitotic response was broad and inclusive, subsequent rounds of proliferative activities relied on fewer cells with enhanced amplifying proliferative capacity. A large majority of clones gave rise to 1–4 cells, however others were able to generate as many as 84. These findings are consistent with experiments performed on endothelial cells isolated from the vessel wall that uncovered distinct proliferative ability of cells over time (Ingram et al., 2005) and a recent report of differential proliferative capacity in endothelial cells dependent upon Sox18 expression (Patel et al., 2017). Importantly our data in vivo cannot distinguish between determination and induction of this functional state. In other words: Is the enhanced proliferative ability of these cells pre-determined or stochastically induced?

To start tackling this question, single cell RNAseq analysis of aortae from young and old mice revealed that amongst the transcription factors involved in regeneration, Atf3 was significantly reduced in aging mice. Atf3 is a highly conserved transcription factor that is known for its response to a range of stress signals and it has been shown to affect both metabolic and immune homeostasis (Aung et al., 2013). Indeed, we found that replication of aortic endothelial cells required Atf3 and that removal of this transcription factor significantly impaired the ability of young mice to regenerate. Essentially removal of Atf3 aged the endothelium, as the response of young mice in the absence of Atf3 resembled the poor regenerative capacity noted in old wild-type mice. Interestingly, deletion of Atf3 has been shown to impair regeneration of peripheral nerves in a manner that appears to be dependent of Jun, STAT3 and Smad1 (Fagoe et al., 2015, Gey et al., 2016). More recently, it has been implicated in intestinal regeneration through the regulation of JNK signaling (Zhou et al., 2017).

Overall these findings revealed that adult quiescent endothelium of large arteries can rewire its transcriptional program to engage in rapid mitotic activity, in the context of pulsatile and high shear stress forces. Several stress response genes, including Atf3 are required for this process. Interestingly, aged endothelium shows a significant reduction in Atf3, as well as its ability to regenerate. Whether increased Atf3 expression in aged endothelium could increase its ability to regenerate has not been tested. It would be unlikely that a single factor would be sufficient, but it would be certainly exciting to try.

STAR Methods

Contact for Reagent and Resource Sharing

Further information and requests for reagents may be directed to, and will be fulfilled by, the Lead Contact, Luisa Iruela-Arispe (arispe@mcdb.ucla.edu).

Experimental Model and Subject Details

Animal care and use

The University of California Los Angeles Institutional Animal Care and Use Committee approved all animal protocols, and all procedures were performed in accordance with these protocols. All animals were maintained at the UCLA vivarium according to the policies instituted by the American Association for Accreditation of Laboratory Animal Care.

Mouse Lines and Genotyping

Wildtype C57BL6/J mice were sourced from UCLA’s internal breeding facilities. VE- Cadherin Cre-ERT2 (Sörensen et al., 2009) and lox-stop-lox-tdTomato reporter mice were a kind gift from Prof. Ann Zovein of UCSF. Rainbow reporter transgenic mice (Rinkevich et al., 2011) were contributed by Prof. Reza Ardehali. FoxM1 lox/lox mouse was a gift from Vladimir Kalinichenko (Cincinnati Children’s Hospital), Myc lox/lox mouse was purchased from Jackson’s laboratories, and Atf3 knock out mouse was a gift from Tsonwin Hai (Ohio State University). All transgenic lines were maintained on a C57BL6/J background.

Male and female animals were used in approximately equal numbers for all experiments except for RNA-seq and parabiosis experiments. For RNA-Seq, exclusively male animals were used to minimize sex differences in gene expression. For parabiosis, exclusively female animals were used in order to minimize aggression between members of parabiosed pairs.

Method Details

Aortic Injury

To create aortic injuries, we surgically dissected surrounding tissue from the abdominal aorta of mice, then placed a strong vascular clamp 800 µm in width on the vessel for sixty seconds (Video S1 and Figure S1 both related to Figure 1, also see Shirali et al., 2016) before closing the abdomen. The pressure from the clamp results in loss of the endothelial lining at the clamping location. This technique was used for experiments where a precise, reproducible wound was desired (Figs 1, 2, 5, 6 and 7). For the study of cell type contributions and regenerative clonal dynamics, we extended the clamping technique to remove larger areas of endothelial lining 2.5–3.0 mm in width by clamping overlapping regions spanning the length of the abdominal aorta from renal arteries to iliac bifurcation (Figures 3 and 4).

Parabiosis Procedure

Parabiosis was performed as described previously (Kamran et al., 2013). Specifically, under aseptic conditions and with continuous isoflurane anesthesia, matching skin incisions are made from the olecranon to the knee joint. Once the corresponding olecranon and knee joints are sutured together, the dorsal and ventral skins are then closed by wound clips (7 mm, CELLPOINT SCIENTIFIC INC). Blood chimerism was confirmed by flow cytometry.

Flow Cytometry to Assess Parabiosis Chimerism

At the time of aorta harvest, 500 uL whole blood was collected in EDTA-coated microtubes via right ventricle puncture with a 25G syringe. Red blood cells were then lysed using a lysis buffer and stained with an Alexafluor-647 conjugated monoclonal rat anti-CD45 antibody (BD CAT#565465) in FACS buffer at manufacturer’s recommended concentration to allow for specific identification of leukocytes. One liter of red blood cell lysis buffer was composed of 82.9g ammonium chloride, 10g potassium bicarbonate, 0.37 grams EDTA disodium salt, and water to 1L with pH adjusted to 7.2. A working solution was diluted 1:10 in water for use the same day. FACs buffer was composed of calcium- and magnesium- free phosphate buffered saline solution with 2% fetal bovine serum and 0.1% sodium azide added.

Aortic Dissection and Fixation

Aortae were fixed in place by perfusing anaesthetized mice with a solution of 4% polymerized paraformaldehyde dissolved in PBS through the left ventricle of the heart for ten minutes. Following perfusion, we removed the aorta from the abdomen, then splayed open the tubular structure as a flat imaging plane using a single cut through the dorsal portion of the segment wall. We then pinned flattened aortic segments lumen-side up atop a 35mm silicone-coated dish, and post-fixed in 4% PFA solution at 4C for one hour with gentle agitation.

Aortic Immunostaining

Staining was performed following immunocytochemistry protocols in a 1.5mL volume sufficient to cover pinned aortae in the dish. Fixed aortic segments were washed three times in 1X HBSS then incubated in blocking buffer containing 3% normal donkey serum and permeabilizing detergent for an hour at room temperature. A solution containing the primary antibodies diluted 1:200 in blocking buffer was then applied and allowed to incubate overnight at 4C. The following day, the aorta was washed three times with 1X HBSS, then incubated with secondary antibodies diluted 1:400 in blocking buffer and a 10mg/mL DAPI solution diluted 1:1000 in the same buffer. After a final set of three washes, the aorta was mounted on glass slides.

Aortic H&E and Verhoeff Stainings

Dissected aortic segments were mounted in HistoGel, then passed to the UCLA Clinical Pathology Core Facility for staining following standard protocols.

Aortic En Face Imaging and Image Processing

Stained aortae were mounted on glass slides with lumen facing the cover slip in Prolong Gold with DAPI (ThermoFisher CAT#P36931) or without DAPI (ThermoFisher CAT#P36930) mounting medium. Aortae were imaged using an LSM880 confocal microscope (ZEISS) with imaging settings optimized per experiment. To image the large, wavy surface we utilized z-stack and tile scan features, stitching the resulting tiles into a single large image (ZEN 2.0 Black software, ZEISS).

Aortic tdTomato Labeling

Homozygous tdTomato reporter mice were crossed with homozygous VE-Cadherin CRE- ERT2 mice. The resulting F1 compound heterozygotes were induced to recombine at high efficiency by intraperitoneal administration of 2 mg tamoxifen free base dissolved in sunflower seed oil (Sigma-Aldrich CAT#S5007) once daily for five days. One week following recombination, the mice underwent aortic injury. One month following injury the aortae were harvested, stained for endothelial marker ERG and injury marker Fibrin, and then imaged. This was shown in Video S2. Cellular Evaluation of tdTomato Cells Post- Regeneration, related to Figure 3. The penetration of tdTomato positive cells in the aorta ranged from 84 to 99%, thus the analysis in the regeneration area must be normalized to the re-penetration, as per tdTomato expression, in the non-regenerated area of the same mouse.

Aortic Rainbow Labeling

Homozygous Rainbow reporter mice were crossed with homozygous VE-Cadherin^CRE-ERT2 mice. The resulting F₁ compound heterozygotes were induced to recombine at a clonal density by a single intraperitoneal injection of 1 mg tamoxifen free base dissolved in sunflower seed oil (Sigma-Aldrich CAT#S5007). One week following recombination, the mice underwent aortic injury surgery as previously described. One month following injury the aortae were harvested, stained for nuclear endothelial marker Erg, and then imaged.

RNA extraction and RNA sequencing from regenerating aortae

We extracted RNA by a lumenal flush method (Briot et al., 2014) and depicted in Figure 5. Specifically, segments of the aorta were dissected, and using a 30G syringe 50uL of RLT buffer was flowed through the lumen of the segment and into a collecting tube, then processed according to the manufacturer’s standard protocol (Qiagen RNEasy Micro Kit, CAT#74004). Approximately 20ng/sample of total RNA was obtained. After verifying RNA integrity using a BioAnalyzer 2100 (Agilent), library preparation and sequencing were performed by the UCLA Genomics Core Facility using the NuGEN Oviation Ultralow Library Prep System (NuGEN CAT#0344) following manufacturer’s protocols; unpaired 50bp reads were obtained at a depth of 40 million reads/sample using an Illumina HiSeq 3000. The enrichment of transcriptome for specific cell type (endothelial, smooth muscle) and heat map of cell cycle and transcription factors is shown in Figure S4 (related to Figure 5). Changes in arterial cell identity during regeneration is shown in Figure S5 (related to Figure 5). The normalized counts are described in Table S1: Transcriptional Changes during Regeneration, related to Figure 5.

Single cell RNAseq

Isolation of endothelial cells from the aorta was performed by first dissecting the aortae free of adventitia and exposing the endothelium to the surface. After several washes with versene buffer, the endothelium was bathed in 1X trypsin and incubated for 10min at 37C. After this time and under the dissecting microscope the surface of the endothelium was gently removed with a small scalpel (microfeather™) and repeating pipetting to aid in the dissociation of cells. The trypsin solution containing released cells was added to a tube containing 0.04% BSA and 0.01% FCS. Schematics of the endothelial cell isolation process, as well as monitoring of the efficiency of endothelial removal is shown in Figure S6, related to Figure 6.

For the generation of single-cell GEMs a suspension of 8700 cells were loaded on a Chromium Single Cell Instrument (10x Genomics) with an estimated targeted cell recovery of ~5000 cells. Single-cell RNAseq libraries were prepared using the Chromium Single Cell 3’ Library & Gel Bead Kit v2 (10x Genomics). Sequencing was performed on Illumina HiSeq2500, and the digital expression matrix was generated by de-multiplexing, barcode processing and gene UMI (unique molecular index) counting using the Cell Ranger v2.0 pipeline (10x Genomics). Data quality, total number of cell clusters, and transcriptional analysis using Seurat have been provided in Figure S7, related to Figure 6.

Quantification and Statistical Analysis

RNA-Seq Data Analysis

Library preparation and sequencing were performed by the UCLA Genomics Core Facility using, from whom we received de-multiplexed reads in the form of FASTQ files. We proceeded to analyze the data using the Galaxy Server interface with the HISAT2 algorithmn and mapped reads to ENSEMBL annotated genome version. Again, through Galaxy, we determined the raw counts of reads aligning with each annotated transcript using ht-seq, then excluded transcripts that do not code for proteins again making use of ENSEMBL annotations. Having obtained raw counts of transcripts for protein-coding genes, we used the Bioconductor suite in R Studio to interface with DESeq2 software. Using DESeq2, we calculated normalized counts for each transcript and performed pairwise comparison to determine statistically significant differences in transcript levels with an FDR <=.01 (results shared in Table S1). Further analysis was performed using a literature-based candidate approach, as described in the main text.

Single cell RNA-Seq Data Analysis

BCL files were extracted from the sequencer and used as input for the Cell Ranger pipeline to generate the digital expression matrix. The matrix was analyzed using Seurat, an R package designed for single cell RNA sequencing. Specifically, cells were first filtered to have at least 1000 UMIs, at least 100 genes and at most 10% mitochondrial gene expression. The filtered matrix was normalized using the Seurat function NormalizeData. Variable genes were found using the Seurat function FindVariableGenes. The matrix was scaled to regress out the sequencing depth for each cell. Variable genes that had been previously identified were used in PCA to reduce the dimensions of the data. Following this, 15 PCs were used in tSNE to further reduce the dimensions to two. The same 15 PCs were also used to group the cells into different clusters. Next, markers were found for each cluster and used to define the cell types. Subsequently, endothelial cells were extracted and used for further analysis. Atf3 expression was analyzed, and was later used to cluster the endothelial cells into an Atf3 expressing group and Atf3 non- expressing group. Afterwards, markers were detected for the Atf3 expression group.

For the comparison between young and old endothelial cells, canonical correlation analysis (CCA) using the Seurat package was used to integrate the two samples. To make the number of cells consistent, only cells with high sequencing depth from old aorta were retained to match the cell number from young aorta. 10 canonical correlation vectors were used to align the CCA subspace, and tSNE was used to reduce the dimensions to two. Markers were found for the young and old aorta endothelial cells.

Aortic Wound Area and Endothelialization Rate Quantification

Aortic wound area was determined by manually tracing fibrin staining (Figure 1) on maximum intensity Z-projections using FIJI software. Areas lacking endothelium were detected by bright fibrin staining and absence of VE-Cadherin positive cells. Areas which had been regenerated were detected by persistence of dim fibrin staining. Linear endothelialization was calculated as the area marked by bright fibrin staining divided by the average width of this area, and the endothelialization rate as the slope of the line connecting the 0–24 and 0–48 timepoints. Beyond these timepoints the upstream and downstream endothelialization fronts have met, making calculation at extended timepoints potentially inaccurate.

Quantification of Endothelial Cell Elongation in Regenerating Aortae

Endothelial cell perimeters were manually traced using a maximum intensity Z-projection of VE-Cadherin stained cell junction using FIJI software. Only cells with clearly distinguishable borders and nuclei (DAPI stain, not shown) were quantified. Based on these perimeters, a best fit ellipse was determined for each cell using FIJI, and the length of the major axis of this ellipse was used as a measure of each cell’s length (Figure S1, related to Figure 1). Statistical comparison was performed using the non-parametric Kolmogorov-Smirnov test in Graphpad Prism software.

Aortic tdTomato Quantification

Quantification of tdTomato labeling frequency was achieved by automated detection of ERG+ endothelial nuclei, followed by classification of each nucleus as tdTomato+ or tdTomato- based on tdTomato fluorescence intensity colocalizing with the nucleus (Video S2. Processing and Quantification of tdTomato Labeled Regenerated Lining). These image processing steps were carried using out using Imaris 8.0.2 (Bitplane). Regenerated areas were identified based upon widened lumen and trace fibrin staining at the sites of clamping.

Aortic Rainbow Quantification

Quantification of Rainbow labeling frequency was achieved by automated detection of ERG+ endothelial nuclei in confocal imaging data, followed by classification of each nucleus as either unlabeled, Cerulean+, mOrange+, or mCherry+ based on fluorescence intensity co-localizing with the nucleus (Video S3. Assessment of Dilute Labeling and Clonality during the Regeneration of Endothelial Lining, related to Figure 4). Regenerated areas were identified based upon altered GFP fluorescence intensity, widened lumen, and apparent clonal expansion at the site of clamping. These image processing steps were carried out using Imaris 8.0.2 (Bitplane). The aortae used in the quantification are shown in Figure S2, related to Figure 4). Furthermore, ability of each specific clone to expand was further quantified using annular plots enabling for direct comparison according to the three colors visible in the recombination (Figure S3, related to Figure 4).

Quantification of Circulation-Derived Endothelial Cells (Parabiosis Experiments)

Quantification of the GFP status of ERG+ cells in regenerated aortae from parabiosed mice was performed by automated detection of ERG+ endothelial cells using confocal data, followed by classification as GFP+ or GFP- based on fluorescence intensity in the GFP channel co-localizing with nuclei. In cases where the automated approach identified possible hits, we further investigated but were unable to find any bona fide circulation- derived endothelial cells. Two such cases that passed automated screening but were manually debunked are shown in Figure 3J. To be even more thorough, we also attempted the converse approach in the GFP- aortae: we detected GFP+ cells and then assessed ERG status. Still no double positive cells were detected (second approach not shown.) These image processing steps were carried out using Imaris 8.0.2 (Bitplane).

Cell Proliferation Experiments in vitro

Isolation of endothelial cells from the aorta was carried out as described for the single cell seq and under sterile conditions. When using tdTomato cells, we employed VE-Cadherin Cre positive / tdTomato reporter mice that were treated with tamoxifen (3X1mg/IP) two weeks prior to isolation. Once isolated cells were plated in 48 wells previously seeded with liver endothelial cells at 80% confluency. Wells were monitored continuously for 5 days using an IncuCyte ZOOM system (Essen BioScience). Movies were extracted (36 per well) and cell divisions evaluated. An example of such movie is provided (Video S4. Differential Proliferative Potential of Endothelial Cells Isolated from the Aortae, related to Figure 6).

The Atf3 reporter lentiviral construct was generated by using a 3.5kb promoter fragment driving eGFP with three nuclear localization signal motifs. Lentiviral particles were generated by transfecting Lenti-X HEK293 cells (Clonentech) with target, VSV-G pseudotype, and delta8.2 packaging plasmids using Lipofectamine 2000 (Invitrogen). Conditioned medium was collected after 48 hours and passed through 0.45 µm filters. ECs were transduced at 50% confluency overnight in EGM-2 medium containing protamine sulfate (4 µg/ml). Empty pRRL was used as control vector. Cultures were then placed on an IncuCyte ZOOM system (Essen BioScience) for visualization and quantification. Expression of eGFP indicated utilization of Atf3 (Video S5. Cells Engaged in Proliferation Express Atf3, related to Figure 7).

Highlights:

• Regeneration of the endothelial lining is mediated by cells flanking the injury

• Endothelial repair does not require circulation of tip/stalk specification

• Cells driving regeneration express a cohort of stress response genes

• Atf3 is required for regeneration of the endothelial lining of large arteries