Mechanisms of bone blood flow regulation in humans

Abstract

Bone is a highly vascularized tissue. However, despite the importance of appropriate circulation for bone health, regulation of bone blood flow remains poorly understood. Invasive animal studies suggest that sympathetic activity plays an important role in bone flow control. However, it remains unknown if bone vasculature evidences robust vasoconstriction in response to sympathoexcitatory stimuli. Here, we characterized bone blood flow in young healthy individuals [n = 13, (four females)] in response to isometric handgrip exercise (IHE) and cold pressor test (CPT). These provide a strong stimulus for active vasoconstriction in the inactive muscle, and perhaps also in the bone. During sustained IHE to fatigue and CPT, we measured blood pressure, whole leg blood flow, and tibial perfusion using near-infrared spectroscopy. Tibia perfusion was determined as oxy- and deoxyhemoglobin. For both stimuli, tibial metabolism remained constant (i.e., no change in deoxyhemoglobin) and thus tibial arterial perfusion was represented by oxyhemoglobin. During IHE, oxyhemoglobin declined (beginning −0.20 ± 1.04 μM; end −1.13 ± 3.71 μM, both P < 0.01) slower than whole leg blood flow (beginning −0.85 ± 1.02 cm/s; end −2.72 ± 1.64 cm/s, both P < 0.01). However, during CPT, both oxyhemoglobin (beginning −0.46 ± 1.43 μM; end −0.60 ± 1.59 μM, both P < 0.01) and whole leg blood flow (beginning −1.52 ± 1.63 cm/s; end −0.69 ± 1.51 cm/s, both P < 0.01) declined with a similar timecourse, even though the magnitudes of decline were smaller than during IHE. These responses are likely due to the different timecourses of sympathetically mediated vasoconstriction in bone and muscle. These results indicate that sympathetic innervation of the bone vasculature serves a functional role in the control of flow in young healthy individuals.

NEW & NOTEWORTHY The current study is the first one to noninvasively investigate control of bone blood perfusion in vivo in humans, on a moment-by-moment basis. Our results indicate that tibial bone vasculature demonstrates active vasoconstriction in response to sympathoexcitatory stimuli in young healthy individuals. Compared with whole leg vasculature, bone vasoconstrictor response seems to be smaller, delayed, and more variable.

INTRODUCTION

The importance of blood supply to bone was recognized as early as in the 16th century, when microscopists described the basic anatomy of bone and its circulation (1). In the 20th century, it became clear that adequate circulation is crucial for nearly all skeletal functions (2–6), such as bone growth (7), fracture healing (8), and bone homeostasis (9). The extensive network of arteries, arterioles, and capillaries that supply the skeleton receives ∼5%–10% of resting cardiac output (10–12) and provides oxygen and essential nutrients critical for bone metabolism. Without sufficient and well-regulated blood flow to meet metabolic demands, bone cannot maintain its integrity. Indeed, decreased flow has been associated with bone loss (13–15), impaired growth (6, 16), delayed fracture healing (17–20), and fracture nonunion. Although this highlights an important link between appropriate bone circulation and bone health, the underlying mechanisms that regulate bone blood flow on a moment-by-moment basis are poorly understood.

Despite the long-recognized importance of vascular supply to the skeleton, bone circulation has proven challenging to study in vivo in both human and animal models, primarily due to the lack of noninvasive techniques for direct measurements of blood content in a highly dense material such as bone. Much of our understanding of blood circulation to bone relies heavily on invasive animal studies; indeed, the gold standard for measurement of bone blood flow is the radioactive microsphere technique, requiring animal euthanization (9). The few available methods to measure bone blood flow in humans (i.e., magnetic resonance imaging, positron emission tomography, laser Doppler flowmetry) are expensive and invasive, requiring injection of contrast agents or radiotracers. More importantly, a major limitation of these methods is poor time resolution, none of them allows rapid measurements, and/or is limited to single-timepoint measurements of bone blood flow. Recently, near-infrared optical systems have emerged as potential solutions for assessing bone blood perfusion noninvasively, continuously, and inexpensively. Few studies employed near-infrared optical systems to assess bone blood flow in humans, but this was predominately early stage proof-of-concept to assess feasibility (21–26). Thus, the key regulators of bone blood flow have not been thoroughly investigated and their dynamic effects remain unexamined.

Near-infrared spectroscopy (NIRS) can provide a better avenue to understand the physiology of bone blood flow regulation. NIR light in the 650- to 950-nm range penetrates biological tissues; light diffuses into the underlying tissue and is absorbed differently by oxy- and deoxyhemoglobin. Taking advantage of these spectral differences, NIRS provides real-time measurements of hemoglobin. NIRS has been primarily used to assess perfusion in soft tissue (27, 28) and has only recently been extended to assess blood perfusion in bone. In recent human studies, NIRS shows reproducible results in controlled conditions (21) and can track dynamic changes in bone perfusion (25, 29). Our previous work tested the efficacy of NIRS to noninvasively detect real-time changes in bone perfusion in the human tibia on a moment-by-moment basis (30). We found that after exercise, healthy adults increase total hemoglobin in the tibia, whereas individuals with spinal cord injury had no detectable changes in tibial total hemoglobin. This work showed the utility of NIRS, but it was not designed to provide insight into blood flow regulation. Thus, the current study employs NIRS to explore key contributors to the control of acute bone blood flow changes in humans.

Sympathetically mediated vasoconstriction is a critical regulator of flow to almost all vascular beds. Yet, even though the bone vasculature is innervated by a rich network of sympathetic nerves, it remains unclear if bone vasculature evidences vasoconstriction in response to sympathoexcitatory stimuli. The extant literature relies primarily on animal models, with the preponderance of data suggesting that sympathetic activation reduces bone blood flow. Exogenous norepinephrine decreases flow to bone (11, 31–34) and, as with other vascular beds, alpha receptors are the primary mediators of smooth muscle contraction (32). However, sympathectomy in dogs may (35) or may not (36) result in increased blood flow to the tibia. In contrast, spinal cord thoracic transection results in increased bone blood flow in rats (37–39). Conversely, electrical stimulation of bone nerve fibers results in vasoconstriction and reduced flow in dogs (40) and rabbits (41). Early-stage work in humans using photo-plethysmography (PPG) suggests that blood perfusion to bone increases in response to release of sympathetic tone via different physiological stimuli (42, 43); however, this work is confounded by the fact that the PPG signal cannot differentiate between arterial and venous flow. Nonetheless, though poorly understood, sympathetic innervation of the bone vasculature appears to play an important role in the control of flow to bone. However, the characteristics of bone blood flow response to sympathetic stimuli remain unknown.

We aimed to examine tibial blood perfusion in response to two pressor stimuli in humans that increase both sympathetically mediated vasoconstriction and systemic pressure. We used the isometric exercise pressor response (i.e., sustained handgrip force at 30% maximal voluntary contraction until fatigue) and cold pressor test (CPT) (i.e., hand immersion in water at 0°C) to generate time-dependent, progressive increases in sympathetic outflow to the vasculature that result in increasing blood pressure. These provide a strong stimulus for active vasoconstriction in the inactive muscle of the leg, and perhaps, in the bone as well (44). Hence, there are two competing drivers of flow during both of these maneuvers—increasing vasoconstriction to decrease flow and increasing pressure to increase flow. As a result of these opposing influences, blood flow to the whole leg remains constant or may slightly increase or decrease (45). During both stimuli, we monitored beat-to-beat blood pressure, whole leg blood flow via Doppler ultrasound, and tibial perfusion via a custom-made NIRS device. We hypothesized that decreases in tibial blood perfusion would be similar to those in the whole leg blood flow in response to both pressor maneuvers in young healthy individuals.

METHODS

Subjects

Able-bodied individuals [n = 13 (4 females), 24 ± 4 yr] were enrolled in the study. All individuals had body mass index below 29.9 (22.55 ± 2.70), were free of cardiovascular and neurological disease, and had no tibial fractures in the past year. All volunteers had a complete medical history and were instructed to refrain from consuming caffeine and participating in exercise 24 h before testing. The study was approved by the Institutional Review Board at Partners Healthcare and all subjects gave written informed consent before testing.

Protocol and Measurements

Throughout the protocol, subjects were supine positioned and instrumented with a standard 5-lead ECG for continuous heart rate, automated finger photoplethysmography (Finapres, Ohmeda Medical) for beat-to-beat arterial pressure, and oscillometric arm cuff for standard measures of brachial arterial pressure to calibrate finger pressure (Dianmap Dash 2000, GE). Popliteal artery blood flow velocity (i.e., representing whole leg blood flow in the muscle, skin, and bone) was recorded using a 4-MHz Doppler probe (Multidop T2, DWL) at the popliteal fossa of the right leg. Although changes in vessel diameter over time would confound the estimate of vessel responses, previous work has shown that femoral and popliteal artery diameters remain unchanged during CPTs (46) and isometric handgrip exercise (IHE) (47). Thus, popliteal blood flow velocity was used as a reliable estimate of blood flow to the whole limb during both stimuli. All signals were digitized at 1 kHz and stored for subsequent offline analysis using commercial hardware and software (ADInstruments, PowerLab). Bone perfusion was assessed using the NIRS system placed mid shaft, on the anterior side of the tibia, in the same leg as the measurement for popliteal artery blood flow. The primary site for active vasoconstriction in response to sympathetic excitatory stimuli is the inactive muscle of the leg. Hence, by investigating flow in both whole limb (via Doppler) and tibial bone (via NIRS) of the same leg, we can identify if the bone has a vasoconstriction response similar to the one in the muscle.

The custom-made NIRS device (Fig. 1) uses a broadband light source and two high-sensitivity spectrometers that yield full NIR spectral information. A white-light tungsten halogen lamp (Fostec DCRII, Dallas, TX) delivers light through a large optical bundle to the skin. Two detector fibers coupled to two spectrometers (Mini-spectrometer TG C9405CB, Hamamatsu Photonics, Japan) are placed at 1 cm and 2 cm distances (center-to-center) from the source. Both spectrometers measure the diffusely reflected light between 650 and 800 nm with 5 nm resolution. The light source and two detectors are attached to a custom-made probe head that is placed on the skin surface, directly over the tibia. Light propagates between the source and the detector in “banana-shaped” sensitivity functions, penetrating at depths corresponding to the distance of separation. Thus, the detector 1 cm away from the source probes the relatively thin (2 mm) skin layer, whereas the detector 2 cm away from the source probes deeper into the underlying bone. Placing the NIRS probe on the anterior side of the tibia, where the bone lies directly under the skin, minimizes the effects of soft tissue and local changes in skin blood flow on the NIRS measurements in bone. The extended modified Beer–Lambert law for a two-layer model is used to determine oxy- and deoxyhemoglobin changes in bone from changes in light absorption. A detailed description of the custom NIRS device and the subsequent offline analysis can be found in our previous work (30).

Figure 1.

Custom near-infrared spectroscopy (NIRS) device for monitoring blood perfusion in tibia. The detector 1 cm away from the source probes only the relatively thin skin layer; the detector 2 cm away from the source probes deeper into the underlying bone. The extended modified Beer–Lambert law for a two-layer model is used to determine oxy- and deoxyhemoglobin changes in bone from changes in light absorption.

Prior to instrumentation, maximal voluntary handgrip force for each volunteer was determined from at least three maximal contractions on a handgrip dynamometer. Following instrumentation, after a 5-min resting baseline, subjects performed sustained IHE at 30% of maximal voluntary handgrip force until fatigue. The target force was displayed on a computer monitor and the subject was continuously encouraged to maintain the target force until exhaustion. The test was ended when the handgrip force dropped >10% below the target for >2 s despite verbal encouragement and the attainment of maximal perceived exertion. Following a 10-min recovery, subjects performed a CPT. After a 5-min resting baseline, the volunteer immersed the right hand in ice water at 0°C for 3 min, followed by 1 min of recovery.

Data Analysis

For both tests, heart rate response was derived from time difference between successive R-wave peaks of the electrocardiogram. Systolic and diastolic blood pressure were derived from maxima and minima of the pressure waveform and mean blood pressure was derived as the sum of 1/3 of systolic blood pressure and 2/3 of diastolic blood pressure. Mean popliteal blood flow velocity characterizing whole leg blood flow was obtained from the area of the Doppler waveform. Tibial bone perfusion (i.e., oxy- and deoxyhemoglobin) was obtained using the algorithm based on the extended modified Beer–Lambert law for a two-layer model from changes in light absorption (30). The NIRS technology does not allow absolute measures of perfusion, but rather relative changes in hemoglobin content (in µM) from one state to another. Thus, we obtained changes in oxy- (ΔHbO₂) and deoxy- (ΔHHb) hemoglobin in the tibia during both IHE and CPT relative to the 20-s resting baseline before the start of each physiological stimulus. Given unchanged metabolism during IHE, deoxyhemoglobin should be virtually zero throughout IHE. To ensure equivalent comparisons between the measurements, we also derived changes in blood pressure (ΔBP), and leg blood flow velocity (ΔLBF), relative to the 20-s resting values before each test. To account for the different durations of IHE to fatigue across subjects, all data were normalized to 100% of the longest test duration. This normalization was not necessary for the CPT as all subjects performed a 3-min test.

For the responses to both physiological stimuli, data from all subjects (normalized to 100% during IHE) were interpolated at 1 Hz. Averages of the population for all variables (i.e., ΔHbO₂, ΔHHb, ΔLBF, and ΔBP) were obtained at each corresponding time point. To characterize the different magnitudes of response over time for the key variables, cumulative sum of differences of blood pressure, popliteal artery blood flow velocity, and tibial oxyhemoglobin were determined for the normalized IHE responses.

Statistics

We used one-sample t test to determine significant changes (i.e., different from zero) from resting baseline at different time points for tibial oxy- and deoxy-hemoglobin, popliteal artery blood flow velocity, and blood pressure across the population. We used paired-sample t test to assess differences in the population between beginning (e.g., 10% IHE duration) and end (e.g., 100% IHE duration) of exercise in the variables of interest. All values are presented as mean ± standard deviation (SD); statistical significance was set at P ≤ 0.05. Data and statistical analyses were performed using custom software written in Matlab (R2018a; MathWorks Inc., Natick, MA).

RESULTS

Peripheral Vascular Response to Sustained IHE to Fatigue

A representative response to sustained IHE to fatigue in one individual is shown in Fig. 2. Popliteal artery flow (LBF) decreased progressively throughout IHE, evidencing active, progressive vasoconstriction in the leg. Similarly, tibial oxyhemoglobin (ΔHbO₂) decreased until the end of exercise, whereas tibial deoxyhemoglobin (ΔHHb) did not change throughout.

Figure 2.

Representative response to sustained isometric handgrip exercise (IHE) in one individual. BP, mean blood pressure; ΔHb, change in tibial hemoglobin from resting baseline; Handgrip target, 30% target force from previously determined maximum voluntary contraction (MVC); LBF, popliteal artery blood flow velocity.

Figure 3 shows the group averaged responses for changes in blood pressure, popliteal artery blood flow, and tibial oxy- and deoxyhemoglobin during IHE. On average, BP increased by 4.63 ± 4.01 mmHg at 10% of exercise and by 46.1 ± 17.9 mmHg at the end of exercise (at 100% duration). LBF decreased progressively throughout IHE, by 0.85 ± 1.02 cm/s at the beginning of exercise (at 10% duration) and by 2.72 ± 1.64 cm/s at the end of exercise (at 100% duration). Similarly, tibial HbO₂ decreased progressively by 0.20 ± 1.04 μM at the beginning of exercise (at 10% duration) and by 1.13 ± 3.71 μM at end of exercise (at 100% duration). Tibial HHb decreased slightly throughout, from 0.06 ± 0.25 μM at the beginning (at 10% duration) to −0.34 ±1.07 μM at the end of exercise (at 100% duration). However, the magnitude of change in deoxyhemoglobin was an order of magnitude smaller than that in oxyhemoglobin, and in fact the average HHb throughout exercise was −0.09 μM, indicating no change in metabolism during sustained IHE to fatigue. Thus, given unchanged metabolism, oxyhemoglobin represented tibial arterial perfusion. All changes in HbO₂, HHb, LBF, and BP at 10% and 100% duration were statistically significant (P < 0.05). Similarly, differences between beginning (10% duration) and end of IHE (100% duration) were statistically significant for all variables (P < 0.01).

Figure 3.

Group average response to sustained isometric handgrip exercise (IHE) to fatigue normalized to longest time duration (100% IHE duration). ΔBP, blood pressure change; ΔHbO₂, tibial oxyhemoglobin change; ΔHHb, tibial deoxyhemoglobin change; ΔLBF, popliteal artery blood flow velocity change. All changes are relative to resting baseline. Data are presented as mean ± SD. n = 13.

Cumulative sum of differences was used to characterize the timecourse and magnitude of response of the key variables during sustained IHE to fatigue. Thus, Fig. 4 shows the group average cumulative sum of differences over time for blood pressure, popliteal artery blood flow velocity, and tibial oxyhemoglobin. Blood pressure demonstrated a rapid, exponential increase during IHE, whereas both leg popliteal blood flow and tibial oxyhemoglobin declined progressively, evidencing active vasoconstriction in the leg. However, the decline in tibial oxyhemoglobin is smaller, delayed, and more variable than in the whole leg blood flow. Moreover, the magnitude of blood pressure increase was 10 times higher compared with that of tibial oxyhemoglobin and whole leg blood flow decreases.

Figure 4.

Average cumulative sum of differences of blood pressure (ΔBP), popliteal artery blood flow velocity (ΔLBF), and tibial oxyhemoglobin (ΔHbO₂) during sustained isometric handgrip exercise (IHE) to fatigue. Please note that blood pressure has four times the y-axis range compared with tibial oxyhemoglobin and popliteal artery blood flow velocity. Data are presented as mean ± SD. n = 13.

Peripheral Vascular Response to CPT

The averaged responses for blood pressure, whole limb blood flow, and tibial oxy- and deoxyhemoglobin during CPT are shown in Fig. 5. On average, BP did not increase during the first 0.5 min and increased only modestly throughout CPT, reaching a plateau of 17.5 ± 10.3 mmHg at the end of the task. LBF decreased by 1.52 ± 1.63 cm/s during the first 0.5 min, reaching a decline of 0.69 ± 1.51 cm/s at the end of CPT. Tibial HbO₂ decreased by 0.46 ± 1.43 μM in the first 0.5 min, with only a slight decrease throughout the remainder of CPT, reching a decline of 0.60 ± 1.59 μM by the end. Tibial HHb declined slightly throughout, from 0.024 ± 0.22 μM in the first 0.5 min, to −0.17 ± 0.77 μM at the end of CPT; however, as in the case of IHE, the magnitude in HHb was an order of magnitude smaller than that in HbO₂, and in fact the average HHb throughout exercise was −0.02 μM, indicating no change in metabolism during sustained CPT. Thus, as in the case of IHE, given unchanged metabolism, oxyhemoglobin represents tibial arterial perfusion. All changes in HbO₂, HHb, LBF, and BP at the different time points were statistically significant (P < 0.01). Similarly, all differences between beginning and end of CPT for all variables were statistically significant (P < 0.01).

Figure 5.

Group average response to cold pressor test (CPT). ΔBP, blood pressure change; ΔHbO₂, tibial oxyhemoglobin change; ΔHHb, deoxyhemoglobin change; ΔLBF, popliteal artery blood flow velocity change. All changes are relative to resting baseline. Data are presented as mean ± SD. n = 13.

Given smaller responses of all the key variables during CPT compared with IHE, the cumulative sum of differences is not presented for CPT as it provides no additional information on either the timecourse or magnitude of response for blood pressure, popliteal artery blood flow velocity, and/or tibial oxyhemoglobin. Nonetheless, to be noted that both leg popliteal blood flow and tibial oxyhemoglobin declined quickly at the start of CPT, maintaining a similar decrease throughout the task, indicating active vasoconstriction in the leg.

DISCUSSION

Adequate vascularization of bone tissue is essential for ensuring a healthy and functional skeletal system. However, fundamental aspects of blood flow regulation to bone remain poorly understood, especially on a moment-by-moment basis. Thus, this work explored the role of a key contributor to regulation of beat-to-beat bone blood perfusion, namely the sympathetic nervous system. We investigated tibial blood perfusion using NIRS in young healthy individuals in response to two pressure stimuli that increase both sympathetic outflow to the vasculature and blood pressure. The decrease in tibial hemoglobin in response to increased sympathetic outflow to the vasculature during both IHE and CPT indicates that in young healthy individuals, bone demonstrates sympathetically mediated vasoconstriction. However, the response is smaller, delayed, and more variable than in the whole leg vasculature. Also, our results indicate that bone vasculature is not simply a pressure passive system. Similar to other vascular beds, sympathetic innervation of the bone vasculature serves a functional purpose in the control of flow. If that were not the case, tibial oxyhemoglobin would have simply followed blood pressure, despite the active vasoconstriction of the inactive muscle of the leg.

For both stimuli, we assumed that leg metabolism remains constant, which was evidenced by virtually no change in deoxyhemoglobin throughout IHE and CPT. Thus, tibial arterial perfusion was represented primarily by the change in oxyhemoglobin. During IHE, the decline in oxyhemoglobin is slower than the decline in whole leg blood flow. Contrary, during CPT, both the decline in oxyhemoglobin and whole leg blood flow followed a similar timecourse, even though the magnitudes of decline were smaller than during IHE. These responses are likely due to the different timecourses of sympathetically mediated vasoconstriction. The increase in sympathetic outflow in response to IHE typically exhibits a 1- to 2-min delay (∼35% duration) following the onset of exercise, with slowly increasing vasoconstriction until reaching fatigue (48). In contrast, sympathetic outflow increases almost immediately during CPT, reaching peak vasoconstriction during the second minute of hand immersion (49).

The current results indicate that sympathetically mediated vasoconstriction plays an important role in the regulation of blood flow to bone in humans. Although animal data have previously shown that sympathetic activation reduces blood flow to bone (31), there are currently no studies that have investigated moment-to-moment regulation in vivo in both animals and humans. Recent studies using PPG in humans suggest that blood perfusion in the bone microvasculature increases in response to release of sympathetic tone via different physiological stimuli (42, 43, 50, 51) and that myogenic control is the predominant bone blood perfusion mechanism in response to changes in external pressure (51). However, it is unclear if the changes observed are representative of the bone microvasculature. The pulsatile component of the PPG signal used to monitor local tissue perfusion is unlikely to represent only capillary blood content, as there is no pulsatile flow in the capillaries. Thus, the PPG measurement captures blood content in other vessels of the microcirculation such as the arterioles and the venules; but it is unclear how to differentiate between arterial and venous blood flow. Given the emitting light at the isosbestic point, where both oxy- and deoxyhemoglobin have the same absorption value, one could not discriminate between the two. Thus, any change in PPG signal could be due to an increase in blood content on either the arterial or venous side. The custom-made NIRS device used in this study has a broadband light source to yield full NIR spectral information, as opposed to measurements at only two wavelengths, to ensure that oxy- and deoxyhemoglobin values can be obtained. In addition, using two detectors as opposed to one minimizes the influence of the skin tissue, which is highly perfused and may cause significant artifacts.

One limitation of the NIRS technology is that it provides only relative changes in perfusion from one state to another. The intensity of the detected light measured across the tibia cannot be used for an absolute measure of perfusion. However, as blood content increases, there is a proportional increase in light absorption (and vice versa). Thus, the hemoglobin change detected reflects the relative change in perfusion. Moreover, given the fixed distance between the light source and the detectors, the hemoglobin content is representative of the same sample volume, regardless of leg size.

An important consideration for the current study and future directions is the assessment of sympathetic nerve activity to the tibial vasculature. Microneurographic measurements of the common peroneal nerve that innervates primarily the fibula, but also the tibialis anterior that overlies the tibia, would likely reflect sympathetic outflow to both muscle and bone of the lower leg. This would allow a quantification of the vasoconstriction effect on the bone, as well as identifying possible delays in vasoconstriction response.

Conclusions

Despite immerging interest in the bone vasculature, our understanding of the mechanisms that regulate blood circulation to bone is limited. To our knowledge, the current study is the first one to investigate the control of bone blood content in vivo in humans, on a moment-by-moment basis. Our results indicate that tibial bone vasculature demonstrates active vasoconstriction in response to sympathoexcitatory stimuli in young healthy individuals. As in other vascular beds, sympathetic outflow appears to play an important role in the regulation of blood content to bone. However, compared with whole leg vasculature, bone vasoconstrictor response seems to be smaller, delayed, and more variable. Future studies will quantify sympathetic outflow to the leg vasculature and its effect on bone vasoconstriction response.

GRANTS

This work was supported by NIH 1R21AR074054 and Paralyzed Veterans of America Research Foundation (to A.E. Draghici).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

J.A.T. and A.E.D. conceived and designed research; A.E.D. performed experiments; A.E.D. analyzed data; A.E.D. and J.A.T. interpreted results of experiments; A.E.D. prepared figures; A.E.D. drafted manuscript; A.E.D. and J.A.T. edited and revised manuscript; A.E.D. and J.A.T. approved final version of manuscript.