Allometric scaling of endothelium-dependent vasodilation: Brachial artery flow-mediated dilation coming of age?

Current efforts at reducing the burden of cardiovascular disease (CVD) extend beyond management of end-stage sequelae, to early recognition of ‘intermediate’ phenotypes. Vascular endothelial dysfunction is an early and convergent step in atherosclerosis, and its presence implies disruption of a multitude of endothelium-dependent physiologic processes. Eventually, dysfunctional endothelial cells promote a proinflammatory and procoagulatory vascular milieu that hastens clinical onset and progression of disease. As these abnormalities precede morphologic changes, assessment of endothelial function can therefore gauge the health of the vasculature, explore underlying mechanisms of CVD, and serve as a clinical endpoint. ¹

Although a complex cascade of events underlies the evolution of endothelial dysfunction, a major abnormality that ensues early with increasing burden of proatherogenic injury is impairment of endothelium-dependent vasodilation. Normally, vascular endothelial cells respond to mounting shear stress by local release of nitric oxide (NO) and vascular smooth muscle relaxation. This can be construed as a homeostatic mechanism that regulates luminal stress incurred by the vascular wall; it is evident in the microvasculature as well as the conduit vessels. Other vasoactive compounds are also secreted by the endothelium with increased shear stress, including prostaglandins and endothelium-derived hyperpolarizing factors.^2,3 However, when specific testing conditions are applied, the vasodilator response is largely NO-mediated. This is particularly true when the shear stimulus is acutely large and transient (i.e. reactive hyperemia following vessel occlusion). In contrast, NO-independent mechanisms are known to contribute significantly to the responses observed with gradual or persistent increases in shear stimulus (i.e. exercise, warming). ⁴

Early approaches of assessing endothelial function in humans tested the epicardial coronary arteries and involved infusion of endothelium-dependent vasodilators. This is considered the ‘gold standard’ measurement, and preserved coronary endothelial function is defined by a vasodilatory response to intracoronary acetylcholine. Conversely, dysfunctional endothelial cells release less NO and enable the direct actions of acetylcholine on the vascular wall. As a result, intracoronary acetylcholine causes contraction of muscle cells that is not opposed by sufficient NO availability. This ‘paradoxical’ vasoconstriction is the hallmark of coronary endothelial dysfunction. ⁵

Vasodilator responses are similarly elicited in the peripheral vasculature by invasive techniques and were similar to those observed in the coronary vessels. Together with the inherent disadvantages of invasive testing, this led to the development of several non-invasive techniques. Instead of catheter infusion of vasoactive compounds, a fixed ischemic stimuli generates physiologic increases in shear stress, which is commonly achieved by suprasystolic inflation of a cuff placed on the upper extremity. Doppler sonography is utilized for non-invasive measurement of the arterial diameter and flow velocity at baseline, and after cuff deflation. This approach quantifies the vasodilator response and can also estimate the hyperemic shear stress (viscosity * shear rate) by using its surrogate measure: the shear rate (ratio of flow velocity/lumen diameter). This is because blood viscosity is not readily measured in clinical studies and is simplistically assumed to be constant across the arterial tree, and uniform in its contribution to shear stress amongst individuals.

Over the past two decades, brachial artery flow-mediated dilation, or FMD, has been reported as an index of endothelial function in thousands of clinical studies. Early on, it was shown that endothelial cell NO-synthase inhibitors abrogate the majority of brachial FMD, indicating that it is an NO-dependent phenomenon. The popularity of this approach is also related to the impairment of brachial FMD in sub-clinical disease, thus potentially serving as a reversible phenotype. ⁶ Aside from its non-invasiveness and relative low costs, FMD is highly reproducible in specialized laboratories despite its intrinsic operator dependence. ⁷ On the other hand, consensus regarding the best mathematical approaches of presenting the FMD response, as well as the technical methodology of the measurement procedure, remains controversial.

Whether routine application of the FMD response is clinically useful remains to be fully demonstrated and may be hindered, at least partly, by methodological heterogeneity between research laboratories. Nevertheless, there is little controversy that FMD declines with age and is lower in patients with traditional CVD risk factors. Progressive FMD worsening accompanies increasing risk burden, indicating that FMD may reflect both the burden of risk and duration of exposure. Notably, the predictive utility of FMD has been demonstrated in large population studies and was shown to improve reclassification of risk beyond the Framingham risk score. ⁸ Patients with persistent impairment or progressive decline in FMD similarly endure worsened CVD outcomes, and those with improved values fair better. ⁹

In clinical settings, FMD is as accurate as electrocardiography in identifying high-risk obstructive coronary disease in patients undergoing exercise stress testing, and is proposed as a sensitive tool for detecting clinically silent hypertension that was otherwise only evident on ambulatory monitoring.^10,11 Moreover, effective treatment modalities of atherosclerotic CVD, such as statin therapy and exercise, are now known to improve FMD. All these findings provide compelling evidence for possible clinical utilities of FMD in routine cardiovascular practice, including its incorporation in risk assessment, identification of ‘subclinical’ vascular disease phenotypes and utility as a short- and medium-term therapeutic endpoint.

Surprisingly, the protocols for measurement of the FMD response continue to vary widely. ⁷ This effectively precluded meaningful comparison or replication of results reported by different groups and limited the ability of establishing normative values. Differences that directly impact the magnitude of FMD include the location of the cuff (forearm versus upper arm), effectively dictating whether the location of the imaged arterial segment is proximal (brachial) or distal (radial) to the occlusion. Under the latter condition, additional mechanisms are thought to be recruited that are NO-independent, and responses are larger relative to the vessel diameter. ¹ The duration of cuff occlusion, which determines the magnitude of reactive flow and the underlying stimulus, has also varied between groups. Moreover, the hyperemic diameter is imaged at different time points following cuff deflation. Alternatively, several investigators calculated the area under the curve (diameter versus time), as a more ‘comprehensive’ measure of endothelium-dependent vasoreactivity. Other issues that also affect the FMD measurement procedure include the stereotactic location of the ultrasound probe and the expertise of the operator, as well as the methodology used to detect arterial edges (e.g. manual versus partially automated).

More importantly, while an accurate depiction of endothelial function is the obvious goal of measurement, the confounding effects of the baseline vessel caliber or magnitude of the underlying stimulus were evaluated by variable approaches, or not at all, amongst different laboratories. Thus far, the best representation of the FMD response remains elusive and continues to be the subject of a heated academic discourse.

Most studies report the FMD response as the percentage increase in arterial diameter or FMD%, which is the (D_hyperemia – D_base) / D_base * 100%. Like many indices of clinical measurements, ratiometric scaling attempts to normalize for the population variability in D_base. For example, the FMD value of 10% entails 0.5 mm vasodilation in an artery measuring 5 mm but only 0.25 mm in another measuring 2.5 mm, and so forth. As such, presenting the FMD response as the ratiometric index FMD% inherently assumes proportional and uniform vasodilatory capacity across different caliber vessels. This assumption is not supported by early observations that FMD% inversely and consistently correlates with D_base, which in turn is highly dependent on body size, measures larger in men and increases with age and the burden of traditional CVD risk. In fact, a larger D_base predicts progression of disease and future outcomes irrespective of the FMD response.^8,12

Similarly, most structural and functional cardiovascular measurements scale with height and body habitus. However, correction for body size is not routinely practiced, including in the assessment of clinically pivotal measurements, such as the aortic arch diameter or the structure of the left ventricle. Even when reported per body size unit, interpretation of several measurements after ratiometric scaling is mathematically problematic. For instance, division of the cardiac output and stroke volume by the body surface area assumes linearity between volume, a three-dimensional measure (e.g. cm³), and a two-dimensional indicator of body size (e.g. m²), which is clearly violated and requires multi-dimensional analysis for proper correction. While scaling in essence attempts to reach size-independent clinical parameters, many ‘corrected’ cardiac and vascular measures continue to correlate significantly with body size.

Inappropriate scaling of clinical measurements leads to biased estimation of differences that either ‘over’ or ‘under’ scale a true relationship. In effect, this increases the overlap in measurement values between normal variants and diseased states, and may jeopardize the identification of clinically useful measurements. Depending on the specific situation, the implications range from creating spurious but clinically trivial effects, to misleading medical practice and research efforts. Hence, alternate techniques to ratiometric scaling can be invaluable in better defining normative values of cardiovascular measurements.

Allometry is a biologic principle that refers to the variation of morphometric or physiologic traits in relation to the overall size, or, more liberally, any other morphologic characteristic of an organism. ¹³ An ‘isometric’ relationship would be that of an organ that weighs 5% of total body weight across developmental aging (ontogenetic allometry) and in adults of the same age with differing weight (static allometry), and the scaling exponent, or factor, in both cases is equal to one. Alternatively, hyper- and hypo-allometric traits deviate from isometry and exhibit varied relationships across body size categories, whether between adults of conspecific species or throughout developmental growth.

The report by Atkinson and Batterham in this issue of Vascular Medicine scrutinizes FMD% use as an appropriate reflection of the FMD response, in one of the largest available datasets derived from the Multi-Ethnic Study of Atherosclerosis or MESA. ¹⁴ As a classic candidate for indiscriminate ratiometric scaling, the authors argue that FMD% creates a moderate-to-large spurious dependence between itself and D_base. To demonstrate this notion, randomly generated data for baseline and hyperemic diameter without any physiological link were generated. The same negative dependency of FMD% to baseline diameter was again evident, indicating that this relationship is at least partly a statistical artifact. The authors then demonstrate how error is propagated with the choice of a ratiometric index, effectively amplifying the FMD% coefficient of variation through several arithmetic steps involved in its calculation. These inherent shortcomings overestimate ‘endothelial function’ in smaller caliber vessels, and vice versa, and can bias group effects – thereby clouding what constitutes a clinically relevant response. Overall, the authors assert that FMD% is a bungling, likely fallacious measure of FMD.

In order to accurately assess a response that is D_base-independent and fundamental to the FMD procedure, the authors advocate application of allometric scaling principles. Subsequently, analysis of covariance (ANCOVA) using the log-transformed increase in diameter (D_hyperemia – D_base), or D_diff, scaled against log-transformed D_base, allows derivation of a scaling exponent that ‘normalizes’ D_diff to D_base. The authors use this approach to illustrate biased conclusions related to the use of ratiometric FMD% in MESA, including the effects of age, sex, disease state and the Framingham risk score on the FMD response in the MESA cohort.

The results Atkinson and Batterham report in this issue of Vascular Medicine are of unequivocal importance and are highly relevant to clinical vascular research laboratories. For the first time, allometric correction is applied on a large dataset, and the authors outline a novel approach that addresses a major limitation of FMD measurements: its intimate dependence on D_base. The implications of these findings vary widely and may include revision of existing results, to ushering of a new era of refined endothelial function testing. The authors are also commended for providing a supplementary file that includes a step-by-step guide on how to undertake allometric-ANCOVA in order to derive scaling exponents that are study- and situation-specific.

However, Atkinson and Batterham’s choice of D_base for allometric scaling purposes poses new interpretive challenges. First, scaling of the FMD response by D_base weakened the relationship between FMD and CVD in MESA, resulting in a less clinically useful metric than FMD%. The authors postulate that this may be due to the effects of D_base, which is in and of itself a marker of CVD risk, on FMD% that renders it as a ‘hybrid’ index with improved utility in CVD. This in turn questions the appropriateness of scaling by D_base to start with: while random-generated data show that the FMD%–D_base inverse correlation is, at least partly, a statistical artifact, this certainly does not preclude a physiological link between larger vessels and a low FMD response.

Indeed, worsened CVD outcomes in those with a larger D_base plausibly imply concomitant impairment of endothelium-dependent vasodilation. Chronic increases in blood flow and shear precipitate enlargement of conduit arteries may also simultaneously exhaust endothelial NO release. Both abnormalities are evident in CVD, such as hypertension or heart failure, and were assessed by several testing modalities. Thus, when the FMD response is scaled by D_base in two individuals of exactly the same size, the discrete risk inherent to a larger D_base may be lost and therefore spuriously decrease the correlation between CVD and the FMD response.

Second, the authors argue FMD% cannot be retained as a valid metric of the FMD response, despite being the outcome in thousands of clinical studies. This is because D_base is the denominator of FMD%, which renders it unamenable to allometric-ANCOVA with D_base as a covariate. Given these interpretive complexities, and as FMD% is a ubiquitously used metric, with proven predictive utility in CVD, we wonder why the authors did not scale FMD% or D_diff by other indicators of body size for comparison. It would appear that allometric correction with established indicators used for cardiovascular measurements (e.g. height, body surface area or fat-free mass) may be more empirically and theoretically sound and can also retain FMD% as the outcome measure.

In conclusion, FMD has been widely utilized in clinical research over the past two decades and its prognostic value has been recently demonstrated in large population studies. The work of Atkinson and Batterham reported in this issue of Vascular Medicine refines FMD representation and may usher a new era of its use in clinical practice.