Trans-nodal migration of resident dendritic cells into medullary interfollicular regions initiates immunity to influenza vaccine

To characterize LNDC response after vaccination, CD11c-eYFP C57BL/6 mice (Lindquist et al., 2004; Hickman et al., 2008; Sung et al., 2012; Kastenmüller et al., 2013) were immunized s.c. in the footpad with UV-inactivated influenza A virus strain PR8 (UV-PR8). DCs were tracked by multiphoton intravital microscopy (MP-IVM) of the popliteal LN (PLN) by surgically exposing the node in live, anesthetized mice (Gonzalez et al., 2010). Continuous imaging for 40 min after UV-PR8 vaccination revealed an influx of LNDCs proximal to the collagen capsule (<150 µm; Fig. 1 a and Video 1). Quantitation of cellular trafficking over this period identified a three- to fourfold increase in the number of YFP⁺ DCs within this region (3.23 ± 0.24; P < 0.001), suggesting a rapid repositioning to the periphery of the PLN.

In addition to increased cell number within the LN periphery, LNDCs exhibited extensive morphological changes over the 40-min imaging period (Fig. 1 a, inset). Measured in bulk, DCs within these regions increased in surface area by almost 50% after vaccination and experienced a concomitant increase in volume and decrease in spherical index, a measure of the spherical nature of an object (Fig. 1 c and not depicted). Importantly, the fluorescence intensity of individual DCs did not change over this time period, indicating that the observed phenotypic changes were not caused by changes in YFP expression. Together, these data suggest an unexpected accumulation and activation of resident DCs in the PLN periphery immediately after vaccination.

To address the origin of the accumulating DCs, in vivo cellular tracking was applied to the live imaging model. By tracking individual DCs, it was observed that rather than infiltrating from an outside source, the cells originated from the interior of the PLN (>150 µm from the collagen capsule), which is beyond the imaging depth threshold for our imaging system (Fig. 1 b and Video 1). Pretreatment of mice with CD62L blocking antibody or local administration of Pertussis toxin, two approaches which limit influx of leukocytes from the vasculature or lymphatics, had negligible effects on DC accumulation (not depicted). These results provide additional evidence that the activated DCs originated within the PLN before vaccination and confirmed their identity as LNDCs.

To assess the overall movement of LNDCs after vaccination, 50-µm serial cryosections of PLNs were imaged by multiphoton microscopy (MPM) and serially reconstructed for analysis of whole PLNs. Similar approaches were reported by Grigorova et al. (2010) using confocal microscopy for partial PLN imaging. By in vivo labeling medullary macrophages through preinjection of an α-F4/80 mAb into the footpad of CD11c-eYFP reporter mice, the medulla could be outlined and shown to include relatively sparse populations of LNDCs in resting naive LNs (Fig. 1 d). In agreement with the live imaging data, injection of UV-PR8 into the footpad of CD11c-eYFP mice stimulated a visible shift of the LNDC population into the medullary compartment by 60 min after vaccination (Fig. 1 d). Flow cytometric analysis of single cell suspensions of PLN indicated no appreciable increase of LNDCs at these time points. This global repositioning of the LNDCs can be quantified through a shift in the ratio of medulla-occupying versus total LNDCs (Fig. 1 e). Migration data were further verified through MP-IVM, through which the rapid arrival of LNDCs into the medulla could be observed (Video 2). Surprisingly, this change in both LNDC morphology and localization could not be identified after the injection of traditional adjuvants such as alum and MF59 despite extensive inflammation of the PLN, suggesting that this response may be specific toward viral antigen or endosomal TLR signaling.

Recent studies have identified interactions of DCs and T cells outside of the T cell area within IFRs in the stimulation of memory CD8⁺ T cell responses (Hickman et al., 2008; León et al., 2012; Sung et al., 2012). In these studies, central memory T cells were generated after viral infection with the purpose of tracking those cells in the LN after secondary challenge. Although there is debate on the resting location within the LN, it is clear that after secondary challenge, central memory CD8⁺ T cells are rapidly recruited to the IFRs where they undergo activation by antigen-loaded APCs. Furthermore, a study by Hickman et al. (2008) suggested a primary role for these sites in priming CD8 T cell immunity.

We hypothesized that these sites may serve as destinations for migrating LNDCs. To characterize the architectural identity of the IFRs, PLNs were in vivo labeled with antibodies against the lymphatic endothelium (a-Lyve-1) and subcapsular sinus macrophages (a-CD169) and then optically cleared for MP imaging. Projections of whole nodes after optical clearing (Ertürk et al., 2012) identified long extensions of the medulla that protruded extensively between B cell follicles and merged with the subcapsular sinus (Video 3). This observation identified two different types of IFR structure: cortical IFRs (cIFRs), which represent chemokine boundaries between the paracortex and B cell follicles, and medullary IFRs (mIFRs), which represent the most afferent connection points between the medulla and the subcapsular sinus (Fig. 2 a). Interestingly, these regions have recently been targeted by Liu et al. (2014) in vaccination attempts, resulting in greatly enhanced T cell priming. Confocal imaging of mIFRs in PBS or UV-PR8–vaccinated PLNs identified YFP⁺ clusters similar to those identified in reconstruction analysis, identifying these sites as destinations for migrating LNDCs (Fig. 2 b). Histological staining confirmed that both CD11b^hi and CD8a⁺ LNDC populations had accumulated in mIFRs within 60 min of vaccination, suggesting a multisubset responsiveness to inactivated influenza (Fig. 2 b).

Afferent lymph flowing through the medullary sinus is constitutively filtered by sinus-lining macrophages (Gray and Cyster, 2012), and this process predicts a natural gradient of viral antigen after vaccination. Thus, it is proposed that the highest antigen concentrations would reside at the tips of these mIFRs (Video 4). To test this possibility, mice were injected s.c. with UV-PR8 labeled with Alexa Fluor 633. Fluorescent PLN reconstructions at various time points identified accumulation of viral antigen within mIFRs over 6 h after vaccination (Fig. 2 c). In contrast, lower levels of virus were retained in the subcapsular sinus and the deeper medullary compartments connecting with the efferent lymphatics. It was hypothesized that these antigen-rich regions might serve as a destination for migrating LNDCs. High-resolution imaging of IFRs bearing dense deposits of viral antigen showed large numbers of infiltrating LNDCs within 60 min of viral arrival at the node. Analysis of LNDCs by flow cytometry confirmed viral capture by both CD11b^hi and CD8a⁺ LNDCs within 30 min after injection (Fig. 2 e), which was verified by MPM identifying viral patches on the surface of LNDCs at later time points (Fig. 2 f, inset).

Previous work by Itano et al. (2003) interrogated the role of mDCs versus LNDCs in CD4⁺ T cell stimulation. In that study, soluble peptide designed for presentation on MHC II was injected s.c., and downstream T cell responses were tracked in the draining LN. The authors observed two waves of peptide presentation, one at 6 h, which resulted in only transitory activation, and one at 18 h, which stimulated a more robust, long-term T cell response. The authors concluded that although resident DCs can activate T cells, the resident cells induce only a limited response with mDCs from the injection site required for robust immunity. To determine whether LNDCs participate in activation of viral-specific CD4⁺ T cells within this study, CD11c-eYFP mice were adoptively transferred with labeled, ova-specific CD4⁺ OT-II T cells 24 h before vaccination with a UV-inactive recombinant strain of PR8 (UV-PR8-OTII), which was engineered to express the OT-II epitope (Thomas et al., 2006).

As predicted, vaccination with UV-PR8-OTII stimulated activation of cognate T cells within 6 h, as indicated by CD69 up-regulation (Fig. 3 a). Although individual DC–T cell contacts could be observed as early as 6 h, extensive interaction between LNDCs and CD4⁺ OT-II T cells was observed near IFRs by 12 h after vaccination (Fig. 3 b). Additionally, small clusters of viral-specific T cells could be identified surrounding LNDCs in these regions. To quantitate clustering, labeled, naive OT-II T cells were transferred into CD11c-eYFP recipients, which were subsequently vaccinated with UV-PR8-OTII 12 h before LN harvesting. The LNs were serially imaged for reconstruction, and single plane images were captured from each reconstruction, representing 400 µm of vaccinated versus control LNs. Using histocytometric analysis (Gerner et al., 2012), OT-II T cells were identified within each plane, and each T cell was assessed for the number of T cell “neighbors” present in the image within one T cell diameter. The resulting measure of T cell clustering revealed significant increases in the mean number of T cell neighbors and increases in the absolute number of T cells with multiple neighbors within vaccinated versus nonvaccinated LNs (Fig. 3 c). Significantly, heat mapping the number of T cell neighbors onto the original images identified the border between the paracortex and IFRs as the most densely populated by clustered T cells (Fig. 3 d). Analysis of T cell migration at 24 h after immunization identified discreet clusters at the extremes of the paracortex, with specific colocalization of T cell clusters at the mIFR–cortical junctions (Fig. 3, e and f).

Previously mentioned studies describing central memory CD8 T cell responses within cIFRs have suggested the chemokine CXCR3 as a critical mediator of cellular retention in these regions. The ligands for CXCR3, CXCL9 and CXCL10, are produced by both hematopoietic and stromal cells in response to viral challenge and are critical to CD8 T cell recruitment to these sites (Sung et al., 2012; Kastenmüller et al., 2013). Additionally, Groom et al. (2012) have recently described the up-regulation of CXCR3 as an early first step in the generation of robust Th1 CD4 T cell responses. Thus, we hypothesized that the clustering of CD4⁺ OT-II T cells with LNDCs and virus in the IFR after immunization with UV-PR8-OTII could be CXCR3 dependent. To examine whether CD4⁺ OT-II T cells become activated and differentiate to CXCR3⁺ T cells, LNs were harvested from immunized mice, and CD4⁺ T cells were analyzed by flow cytometry for expression of cell surface markers of activation. Notably, an increase in CD69 expression on CD4⁺ OT-II T cells was observed, followed by a corresponding increase in CD44 (Fig. 4 a and not depicted). By gating on newly activated CD69⁺ cells, increases in CXCR3 expression could be identified as early as 12 h after vaccination and continued to increase over several days (Fig. 4, a and b).

As recent reporting has shown that activated DCs arriving from the periphery express CXCL10 and that direct injection of LPS/Poly I:C can induce CXCL10 expression within 12 h in the draining LN (Groom et al., 2012), it was predicted that LNDCs might express this chemokine after immunization with UV-PR8-OTII. Thus, the release of CXCL10 by activated LNDCs within the IFR could explain the extensive clustering of CD4⁺ T cells observed within the antigen-rich IFRs. To test this possibility, CXCL9/10 reporter mice (REX3; Groom et al., 2012) were vaccinated with UV-PR8, and LNs were harvested after 24 h for analysis by flow cytometry and immunohistochemistry. Flow analysis identified the expression of CXCL10 by both CD11b^hi and CD8a⁺ LNDCs after vaccination (Fig. 4 c), although the CD11b^hi population appeared more robustly responsive. Fluorescent reconstructions of vaccinated REX3 PLNs identified high CXCL10 expression on dendritic-looking cells within mIFRs after vaccination, suggesting LNDC expression (Fig. 4 d). Confocal analysis positively identified these cells as DCs through CD11c expression and confirmed clustering of CXCL10-positive DCs within IFRs by 24 h after vaccination (Fig. 4 e). The timing of this event corresponded with both LNDC migration and formation of T cell clusters.

To confirm the importance of CXCL10 in attracting viral-specific T cells to the IFRs, mice were adoptively transferred with differentially labeled CXCR3^+/+ and CXCR3^−/− OT-II T cells and subsequently immunized with UV-PR8-OTII virus. Characterization of the vaccinated recipients revealed a significant defect in the clustering of CXCR3^−/− OT-II cells at 24 h after vaccination, confirming the importance of CXCR3 in T cell migration (Fig. 4 f). Additionally, flow cytometric analysis of CXCR3^−/− OT-II T cells displayed a profound deficiency in both CD69 activation and CD40L expression (Fig. 4, g and h). These results replicate the study by Groom et al. (2012), which described early activation defects, followed by a deficient Th1 response in the absence of CXCR3. Together, these data suggest a model whereby LNDC activation of CD4⁺ T cells at early time points leads to expression of CXCR3 and contributes to the efficient localization of cognate cells into mIFRs, resulting in efficient early activation.

Although the activation of T cells within 6 h of vaccination suggested a role for LNDCs in early activation of CD4⁺ T cell responses to UV-PR8, it remained unclear whether skin-resident mDCs were required to establish an enduring humoral response. Using a system developed by Itano et al. (2003), the contribution of the mDC population could be effectively removed through resection of the injection site within 30 min after immunization (Kissenpfennig et al., 2005). By administering UV-PR8 intradermally in the ear and removing the ear shortly after administration, the relative contribution of mDC and LNDC populations could be evaluated in the auricular LN (ALN).

Using this approach, C57BL/6 mice were vaccinated with PBS or UV-PR8 intradermally, with half of the vaccinated group undergoing ear removal (here referred to as the van Gogh [vG] group) within 30 min after vaccine administration. Using REX3 reporter mice, expression of CXCL10 by both CD11b^hi and CD8a⁺ in WT or vG-vaccinated animals was assessed 24 h after injection/resection. Interestingly, although CXCL10 expression was still robust in both LNDC populations, suggesting normal responsiveness to vaccination despite a lack of mDCs in the vG group (Fig. 5 a), there was a small decrease in expression by the CD8a⁺ subset, perhaps relating to its known reliance on incoming mDCs from the periphery (Allan et al., 2006).

By transferring CFSE-labeled naive OT-II T cells into B6 recipients and using the vG vaccination system, potential deficiencies in T cell proliferation and activation could be assessed in the absence of mDCs from the injection site. Surprisingly, vG group OT-II T cells proliferated similarly to the WT group (Fig. 5 b) and could be tracked through seven divisions in 60 h. Tracking T cell activation by division status, cells could be observed acquiring both CXCR3 and CD40L as they proliferated in both the WT and vG groups (Fig. 5, c and d). Interestingly, although CD40L expression appeared identical in both groups, there was a slight delay in CXCR3 acquisition between the WT and vG group. These findings may also suggest collaboration with mDCs in reinforcing Th1 immunity development as they arrive from the periphery.

As LNDC and T cell activation appeared to be intact despite the absence of mDCs in the vG model, it was hypothesized that LNDCs may be sufficient to drive downstream protective humoral immunity despite the absence of mDCs. To test this hypothesis, the vG vaccination model was used in C57BL/6 mice alongside PBS-vaccinated controls. ALNs were collected at day 7 after vaccination, and germinal center staining appeared normal in both WT and vG groups, confirming effective T cell help despite the absence of mDCs (Fig. 6 a). Antibody titers collected at day 10 after vaccination showed normal increases in IgM, IgG1, and IgG2b in vG versus WT vaccination groups in comparison with unvaccinated controls, suggesting normal class switching and germinal center development (Fig. 6 b).

To test the functionality of this antibody response, vaccinated, or unvaccinated controls were challenged intratracheally with a lethal dose of live PR8 on day 21 after vaccination. Morbidity was monitored for 9 d after infection, at which point the unvaccinated control group was sacrificed as a result of excessive weight loss. Neither WT nor vG groups experienced significant weight loss over the course of infection, indicating protective immunity had been generated with or without mDC arrival from the injection site (Fig. 6, c and d). Finally, antibody titers were assessed at day 9 and compared with both unvaccinated and uninfected controls. As expected, antibody titers showed extensive class switching in both WT and vG groups, and protective IgG2b responses were identical between groups. It is worth noting that although not required for protection, IgG1 antibody titers were slightly, but insignificantly decreased in the vG group, providing additional evidence for LNDC/mDC collaboration in a vaccination setting. Altogether, these data demonstrate that LNDCs are capable of stimulating CD4⁺ Th1-dependent protective humoral responses to influenza vaccination.

An earlier study identifying capture of lymph-borne inactive influenza virus by LNDCs raised the question of the relative importance of this population in overall humoral immunity (Gonzalez et al., 2010). To gain a broader understanding of LNDCs in response to vaccination with inactive influenza, a whole-node imaging approach was developed, and mice bearing a fluorescent reporter for DCs (CD11c⁺) were vaccinated with UV-PR8. After s.c. vaccination, viral antigen was observed within the draining LN within minutes, as expected. Surprisingly, however, the virus accumulated over the first 6 h within specialized sites identified as mIFRs.

Using fluorescent reconstructions of full PLNs isolated at early time points after vaccination, we observed a major repositioning of LNDCs from the T cell cortex to the mIFR where they acquired viral antigen and became activated. This was unexpected because LNDCs are reported to be relatively sessile at steady-state. The capture of virus by the resident DC was biologically relevant because they became positive for the chemokine CXCL10 and formed clusters with viral-specific CD4⁺ T cells that became activated based on expression of CD69, CD44, and CXCR3. Thus, three lines of evidence are presented that support a role for LNDCs in promoting a humoral response to inactive PR8 independent of mDCs: (1) activation of viral-specific CD4⁺ T cells before expected arrival of skin mDCs, (2) activation of viral-specific CD4⁺ T cells within PLNs in the absence of mDCs, and (3) a normal humoral memory response despite surgical removal of the injection site within 30 min of vaccination. Notably, although these findings clearly demonstrate sufficiency of the resident DC population in establishing humoral immunity, they do not address the requirement of mDCs under conditions where lymph-borne antigen is not readily available.

Although several groups have previously described resident DC populations in skin-draining LNs (Lindquist et al., 2004), few have directly assessed their potential in presenting antigen acquired directly from the lymph. In one study, Itano et al. (2003) describe a two-step process of antigen presentation where resident DCs in the LN were capable of acquisition and presentation of soluble peptides, but skin-resident mDCs were required for effective immunity. In another study, Allenspach et al. (2008) found that in the case of soluble peptide administered with CFA, resident DCs were similarly deficient in T cell presentation. Our findings are not inconsistent with these results, as the robust activation/migration of resident DCs cannot be identified after traditional adjuvant administration (alum or MF59). The profound differences in these responses also raise interesting questions about a potential gap between traditional vaccination approaches and natural response to viral antigen. As medullary macrophages rapidly degrade protein antigen as a part of normal lymphatic filtration, it is important to understand how to maximize antigen exposure to efficient antigen presenters, especially when antigen concentration is highest immediately after vaccination. Data from this study suggest that LNDCs are capable of effectively using antigen when appropriately stimulated and could represent an efficient target for vaccination approaches.

An unexpected observation was the rapid kinetics of LNDC repositioning from the paracortical region to the mIFR in response to localization of viral antigen. Although large deposits of LNDCs were observed in viral-loaded IFRs within hours of vaccination, individual LNDC morphological changes could be observed within 12 min of vaccination. The highly directional migration pattern of LNDCs to the sites of viral accumulation suggests a response to chemotactic mediators released from a prestored source after innate sensing of viral antigen, although this source remains yet unclear.

A recent study has identified CXCR3 and its ligands CXCL9 and CXCL10 as required for migration of CD8 memory T cells into IFRs, where they make contact with both antigen and APCs (Sung et al., 2012). Although the distinction between mIFRs and cIFRs was not made in these studies, it is clear that this chemoattractant is important in various conditions of immune response. In our study, formation of clusters of viral-specific CD4⁺ T cells with LNDCs was significantly reduced in CXCR3^−/− OT-II T cells. Because CXCL10 expression is not restricted to LNDCs, this pathway might provide an efficient mechanism for collaboration between LNDCs and incoming mDCs, whereby early activation of naive CD4 T cells, and concomitant expression of CXCR3, allows activated T cells to more easily find mDCs from the injection site, which also express CXCL10. Additionally, it is possible that mDCs might be able to locate hot spots of immune activation within the LN through their own expression of CXCR3.

In summary, using novel whole-node imaging approaches, we observed a multistep system of immune activation whereby viral antigen, LNDCs, and cognate naive CD4⁺ T cells rapidly assemble in highly organized environments within the draining LNs known to be relevant to vaccine design (Liu et al., 2014). In addition to identifying an unexpected repositioning of LNDCs to specialized sites within the draining LN, these findings clarify the importance of CXCR3 in establishing immunostimulatory microenvironments in three dimensions and establish LNDCs as biologically relevant cell populations, sufficient to drive humoral memory response to an influenza vaccine in the absence of mDCs.

Mice used in this study were bred and housed in standard conditions and used between 6 and 12 wk of age. All experimental protocols were approved through Harvard University’s Institutional Animal Care and Use Committee.

In vivo labeling of the PLN subcapsular sinus and medulla was achieved through injection of Alexa Fluor–conjugated (A488, A568, A633) antibody targeting stromal (a–LYVE-1) or macrophage populations (a-CD169, a-SIGNR1, a-F4/80) s.c. in the footpad 4 h before PLN imaging or tissue harvest.

PLN live imaging was performed through surgical exposure of the PLN in anesthetized mice and MPM. Temperature was monitored using a digital thermometer embedded in the imaging chamber created around the PLN and controlled using a closed-circuit water circulation system (Lindquist et al., 2004; Gonzalez et al., 2010).

Collected organs were fixed in 4% PFA, embedded in OCT, and serially cryosectioned (50 µm). PLN sections were serially imaged by MPM and assembled using Imaris software (Bitplane) to obtain three-dimensional reconstructions. Final analysis was performed using Volocity image analysis software (PerkinElmer).

LNs were dissected and fixed overnight in 4% paraformaldehyde. The sample was then incubated in increasing concentrations of tetrahydrofuran followed by dichloromethane. Finally, the PLN was immersed in BABB (benzyl alcohol/benzyl benzoate, mix 1:2) for final clearing and imaging (Lindquist et al., 2004; Ertürk et al., 2012).

Influenza virus was grown in live chicken eggs, purified over sucrose gradients (Szretter et al., 2006; Gonzalez et al., 2010), and labeled using standard Alexa Fluor–labeling protocol (Invitrogen). 10–20-µl s.c. injections were given in the footpad at 10⁶ PFU.

LNs of OT-II mice were processed using Liberase DH (Sigma-Aldrich) digestion, and T cells were isolated through MACS negative selection (CD11c, CD11b, NK1.1, B220, GR-1, CD69, CD8). T cells were labeled in 10 µm CMTMR for imaging or 5 µM CFSE for imaging and flow cytometric identification. 1 or 5 × 10⁶ cells were adoptively transferred into naive recipients for flow cytometric or imaging experiments, respectively.

Images were processed and analyzed using Volocity imaging software. Histocytometric analysis of reconstructions was performed with CellProfiler (Carpenter et al., 2006; Grigorova et al., 2010), and CellProfiler Analyst (Hickman et al., 2008; Jones et al., 2008; León et al., 2012). Analysis of reconstructions was performed with individual XY imaging planes, which were representative of at least 400 µm of reconstructed LNs.

CXCR3^−/− and WT OT-II cells were isolated as above, differentially labeled, and adoptively transferred into C57BL/6 recipients. PLNs were collected, processed, and serially imaged 24 h after vaccination with UV-PR8-OTII. T cell clusters were identified using a blinded approach, and the percentage of each population within these clusters was measured.

Ear removal was performed 30 min after injection of 10 µl PBS or UV-PR8 s.c. between the ear dermal layers. ELISA analysis of serum was performed through immobilization of UV-PR8 on the plate, addition of serum, and probing for specific binding of IgM, IgG1, or IgG2b.

All statistics/graphical representations of collected data were assembled through Prism (GraphPad Software). Mean and SEM are displayed where applicable. Results from Student’s t tests or Tukey’s post tests are indicated by asterisks in the figures (*, P < 0.05; **, P < 0.005; ***, P < 0.001). One- or two-way ANOVA testing is indicated in the legends.

Video 1 shows LNDC migration in response to influenza vaccination. Video 2 shows that LNDCs infiltrate the medulla after UV-PR8 vaccination. Video 3 shows reconstruction of a PLN. Video 4 shows lymph flow through the PLN.

We thank J. Campbell for providing CCR7-deficient mice, N. Anandasabapathy and S. Jones for critical discussion and review of this manuscript, and H. Leung, E. Carroll, S. Lavoie, and M. Ma for technical support.

This work was supported by the National Institutes of Health (NIH)–National Institute of Allergy and Infectious Diseases (grants 1 P01 AI078897, R37 AI054636, and R01 AI039246 to M.C. Carroll; R01CA069212 to A.D. Luster; RO1 DK074500 and PO1 AI045757 to S.J. Turley; and AI107625 to P.G. Thomas), NIH T32 Training Grant in Transplantation (T32 AI007498 to M.C. Woodruff), the American Lung Association (grant RT-224269-N to C.N. Herndon), and the National Health and Medical Research Council, Australia (fellowship 516791 to J.R. Groom).

The authors declare no competing financial interests.

Author contributions: M.C. Woodruff, B.A. Heesters, and C.N. Herndon performed all experiments and analysis described in the text. P.G. Thomas, J.R. Groom, and A.D. Luster developed critical reagents for the completion of the study. M.C. Woodruff, S.J. Turley, and M.C. Carroll designed this study and developed the manuscript for publication.