Physical and geometric constraints shape the labyrinth-like nasal cavity

The Flow in the Nasal Cavity Is Laminar.

It has been suggested that the flow in the nasal cavity is turbulent (7, 8), since the speeds are high and nasal geometry is complex. Indeed, turbulence can easily be induced in the surrounding air by exhaling heavily, as is apparent on a cold winter day. Inside the nose, turbulent flow would induce additional mixing that improves the heating and humidification of the inhaled air (9), but it also implies a larger resistance to flow. It is thus unclear whether turbulence would be beneficial.

To see whether turbulence occurs inside the nasal cavities of animals of various sizes, we first estimate the mean speed u¯ using experimentally determined scaling relations of respiratory quantities with body mass; see Table 1. In particular, we combine the volumetric flux Q, the volume V of the nasal cavity, and the length of the skull as a proxy for the length L of the cavity, to obtain u¯=QL/V≈0.3m/s⋅(M/kg)0.15±0.06. Since u¯ is much smaller than the speed of sound, the flow is incompressible. For the flow to be turbulent, inertia has to dominate viscous dampening. The ratio of these two effects is quantified by the Reynolds number 𝑅𝑒=u¯Rh/ν, where ν≈1.5⋅10−5m2/s is the kinematic viscosity of air (10) and Rh is the characteristic linear dimension of the flow. In complex geometries, Rh is given by the hydraulic radius Rh=2V/S, where V and S are the volume and surface area of the geometry, respectively. Using the scalings given in Table 1, we find 𝑅𝑒=2LQ/(νS)≈70⋅(M/kg)0.36±0.06, so 𝑅𝑒 increases with body mass. Typically, flows are turbulent when 𝑅𝑒 is about 103, with the precise transition depending weakly on the flow geometry; for example, flow between parallel plates (11) is turbulent at a lower 𝑅𝑒 than in pipes (12). Regardless, this indicates that turbulence occurs only in animals heavier than about 103kg. Indeed, numerical simulations have shown that the narrow geometry of the nasal cavity prevents the development of turbulence in humans (13), dogs (14), and rats (15).

Another ubiquitous feature of nasal airflow is the oscillatory motion caused by natural breathing, which generally induces additional resistance and also limits the humidification and heating of the air. However, this is only important when the characteristic length Rh is smaller than the length (ν/f)1/2 associated with the frequency f of the flow (16). Using the scalings from Table 1, we find Rh(f/ν)1/2≈0.7⋅(M/kg)0.09±0.07, which depends only weakly on the body mass M, so the effect of pulsatility is similar for all animals. We show in SI Appendix that the resistance increases by about 50% compared with steady flow, since Rh is comparable to (f/ν)1/2. Note that, if the frequency f were much higher, the resistance would increase strongly, while a lower frequency would imply a larger tidal volume and thus require a larger lung. The associated trade-off might set the respiratory rate, but, since the pulsatility affects all animals similarly, we can simply analyze steady flow here.

The Optimal Nasal Cavity Has a Uniform Gap Width.

We seek the geometry of the nasal cavity with the lowest resistance to airflow for a given efficiency of heating and humidifying the air under the constraint of a given volumetric flux, length, and cross-sectional area. We thus need to calculate the dependence of the airflow and its physical properties on the geometry of the nasal cavity. Since nasal cavities are typically straight, we first focus on varying the cross-sectional shape.

The flow through the cavity is driven by a pressure difference P generated by the lungs. Since the flow in a straight nasal cavity is laminar, stationary, and incompressible, the velocity field only has a component u in the axial direction, which obeys the Poisson equation∇⊥2u(x,y)=−PηL,[1]with u=0 at the walls (SI Appendix). Here, ∇⊥2 denotes the Laplacian in the cross-sectional plane, L is the length of the cavity, and η≈1.8⋅10−5Pa.s is the dynamic viscosity of air (10). Solving Eq. 1 for u, we obtain the volumetric flux Q=∫udxdy, which scales with P. The resistance K=P/Q is then independent of Q and can be expressed asK=CK⋅ηLA2,[2]where CK is a nondimensional parameter that depends only on the cross-sectional shape; see SI Appendix. We can thus quantify the influence of the cross-sectional shape on the airflow by simply studying its effect on CK.

The heating and humidification properties of the nasal cavity can be quantified by the change in temperature and concentration of water vapor in the air after it flowed through the cavity. Both quantities can be described as a passive scalar c that is transported with the flow, diffuses, and is exchanged with the walls of the cavity. In a stationary state, the scalar c thus obeys0=D∇2c−∂z(uc),[3]where D denotes the diffusivity, and boundary conditions are imposed by the epithelial tissue. For simplicity, we first consider the case where the boundary is kept at body temperature and maximal humidity, which implies a constant scalar value cb at the walls. While flowing through the cavity, the cross-sectionally averaged scalar c¯(z) will thus change from its ambient value c¯(0)=ca to approach cb. The extent of this process at the end of the cavity is quantified by the scalar exchange efficiencyE=ln(c¯(0)−cbc¯(L)−cb),[4]which is larger the closer c¯(L) gets to cb. We show in SI Appendix that c¯(z) can be expressed as c¯(z)=cb+∑nane−z/λn, where λn are length scales that follow from the generalized eigenvalue problem−D∇⊥2cn(x,y)=λn−1u(x,y)cn(x,y),[5]and the coefficients an can be determined from the initial value c¯(0) at the inlet. Note that we here neglected the small axial diffusion term, since it is dominated by advection (u¯L≫D). Moreover, in the simple case of a long cavity, entrance effects can be neglected, and only the mode with the largest λn contributes to E. In this case, we haveE=CE⋅DLQ,[6]where CE=Q/(λ1D) is a nondimensional factor associated with the largest λn. We show in SI Appendix that CE depends only on the cross-sectional shape and captures how the shape affects the scalar exchange.

To find the geometry that has minimal resistance K for a given exchange efficiency E, we determine the respective prefactors CK and CE for several simple shapes; see Fig. 2. CK is lowest for a circular shape, but the associated CE is also minimal. As expected, both CK and CE increase when shapes become narrower for larger aspect ratios, which illustrates the trade-off between low resistance and large exchange efficiency. In fact, for all shapes considered, the ratio CK/CE approaches a constant for large aspect ratios, indicating that both quantities increase proportionally. However, not all shapes perform similarly: The resistance of rectangular shapes is up to 40% lower than that of ellipses with the same exchange efficiency.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Comparison of the relative resistance K/E for different cross-sectional shapes as a function of a prescribed exchange efficiency E. Only the prefactors CK and CE are shown, but the top and right axes display the respective values for human noses with parameters given in Table 2. The considered shapes are a circular pipe (large red circle), a square duct (green square), elliptical ducts (red line, ratio of half-axes indicated at support points), rectangular ducts (green line, ratio of sides indicated at support points), rectangular ducts with rounded corners (violet line), N circular pipes in parallel (blue dots), and numerically optimized shapes (black stars). The resistance of the proper cross-sections is bounded by the value for parallel plates (orange dotted line). Solid lines are quadratic interpolations between the support points (dots) where values were calculated numerically and cross-checked with the literature (32).

To understand what geometric features lead to good performance, we numerically determine how shapes need to be altered to lower the resistance at constant exchange efficiency; see Fig. 3 and SI Appendix for details. This sensitivity analysis suggests that sharp corners and narrow regions are detrimental, as indicated by the large arrows in Fig. 3. However, it does not show clearly why rectangular shapes outperform ellipses. To understand this aspect better, we use the sensitivity analysis to obtain optimal shapes by iterative adaptation as described in SI Appendix. Fig. 3 and SI Appendix, Fig. S2 show that optimal cross-sections are dumbbell-shaped, with a slender midsection. We compare the midsections of different shapes by quantifying the width of the shape perpendicular to the centerline as a function of the distance along the centerline. Fig. 3 shows that this gap width is uniform for the optimal and rectangular shapes, while it varies significantly for the ellipse. Taken together, optimal shapes are thus rounded and posses a uniform gap width.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

(A) Sensitivity to shape perturbations of three cross-sectional shapes. The arrows indicate the magnitude of the decrease of the resistance prefactor CK (arrow length) when the shape is locally perturbed in the normal direction (arrow direction) at fixed area A and scalar exchange efficiency CE; see SI Appendix for numerical details. (B) Width perpendicular to the centerline as a function of position for the three shapes shown in A. This gap width is uniform in the midsection of optimal shapes and rectangles, while it varies significantly in ellipses.

The optimal width of the gap can be estimated from the asymptotic geometry of two parallel plates, which provides a lower bound for the achievable resistance; see Fig. 2. This geometry corresponds to a rectangular duct with the two small sides replaced by unphysical periodic boundary conditions, so the cross-sectional area A is still well defined. The prefactor for the scalar exchange in this geometry is CE=7.54Aℓ−2, where ℓ is the plate separation. Using Eq. 6, we can then solve for the ℓ that results in a given scalar exchange efficiency E,ℓ=2.75DτE,[7]where τ=LA/Q is the time it takes air to cross the cavity. The gap width must thus be similar to the typical distance (Dτ)1/2 the scalar diffuses while passing through the cavity (17).

The result of our theoretical considerations is twofold: First, we qualitatively predict that natural selection favors nasal cavities where the separation between the walls is approximately constant everywhere. Second, we quantitatively predict the optimal gap width, either from the aspect ratio of realistic ducts that lead to a given E or by using Eq. 7 as an approximation.

The Theory Agrees with Experimental Measurements.

Nasal cavities described in the literature are typically narrow and exhibit little variation in gap width (2, 3), which agrees with our theory qualitatively. For a quantitative comparison, we consider geometric measurements of nasal cavities of canid and arctoid carnivorans that have been reconstructed in silico from CT scans (18). The associated scalings of the geometric parameters with body mass are summarized in Table 1. The volumes V of the cavities and the lengths of the skulls scale isometrically, but the surface areas of the cavities exhibit significant positive allometry; that is, in heavier animals, it is larger than expected from geometric scaling.

To test whether our theory can explain the observed data, we calculate the surface area of the optimal geometry of the nasal cavity as a function of the constrained parameters D, A, L, Q, and E. Here, D is the scalar diffusivity, which is either the mass diffusivity of water vapor, Dw=2.5⋅10−5m2/s (10), or the diffusivity of heat at room temperature, Dh=2.2⋅10−5m2/s (19). Since both values are similar, we just consider the slower diffusivity, D=Dh, implying that humidification of the air will be slightly better than its heating. The other parameters could be different for each animal, but, since they have not been measured independently in all animals, we use reported scaling relations with body mass for A, L, and Q; see Table 1. Finally, we consider the scalar exchange efficiency E=1 independent of body mass, implying that temperature and humidity equilibrate with the walls to about 60% in all animals. Indeed, inhaled air is typically heated to about 65% of body temperature (20), and the relative humidity is raised to about 80% (21) in humans.

We first predict the surface area S in the simple geometry of parallel plates, where we can calculate the optimal gap width ℓ explicitly using Eq. 7. Using the scalings described above, we find ℓ≈6mm⋅(M/kg)0.09±0.04. The associated surface area is S≈2V/ℓ, and thus scales as S≈9.8cm2⋅(M/kg)0.87±0.07. Fig. 4A shows that this predicted scaling agrees well with the measured data. Consequently, our simple scaling theory correctly predicts important geometric properties of real nasal cavities. In particular, the positive allometry of the surface area, which is observed in a wide range of animals (5, 22), is a direct consequence of the negative allometry of the optimal gap width ℓ. Note that, if the gap width scaled isometrically (with M1/3), the scalar exchange efficiency E would drop significantly in larger animals, because it scales as E≈ℓ−2. Taken together, our theory shows that the allometric scaling of geometric parameters of the nasal cavity is a consequence of the physics of the airflow and the scalar exchange.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Comparison of the theoretical predictions to experimental data. (A) Surface area S of nasal cavities as a function of body mass M. Shown are experimentally measured respiratory turbinal surface areas in canid and arctoid carnivorans (18) (blue dots) and in humans (orange cross; bars indicate standard deviation; parameters in Table 2). Our predictions (green lines) follow from the optimal gap width of parallel plates given in Eq. 7 (dark green) and a numerical solution based on a rectangular geometry (light green; dashed part indicates square geometry). Here, we assumed E=1 and used the scalings given in Table 1. (Inset) The associated volumes V of the nasal cavities as a function of M together with the scaling given in Table 1. (B) Local gap width, given by the shape diameter function (24), for the standardized nasal cavity (23). (C) Hydraulic radius Rh along the main axis z of the standardized cavity (23) (blue line) compared with the predicted ℓ given by Eq. 7 (green line), which was calculated for E=1 and the parameters given in Table 2. (Inset) The cross-sectional area A of the standardized cavity as a function of z.

So far, we considered the idealized geometry of parallel plates, which contains unphysical periodic boundary conditions. For large aspect ratios, this is a good approximation of the more realistic shape of a rectangular duct, but we showed above that the gap width does not scale isometrically and the aspect ratio thus varies with body mass. In particular, smaller animals will have aspect ratios closer to unity, and the scalings derived from the parallel plate model are not accurate in this case; see Fig. 4A. To correct this, we numerically determine the rectangular shape with a given cross-sectional area A that leads to the exchange efficiency E=1. The light-green line in Fig. 4A shows the associated surface area as a function of body mass, which now cannot be expressed as a simple scaling law. Note that this correction is insignificant for large animals, which confirms that they have nasal cavities with high aspect ratio where the parallel plates model is accurate. Conversely, there are large deviations for small animals, where the side walls contribute to the surface area significantly. In fact, we find that the exchange efficiency exceeds 1 for small animals even in the case of a square geometry (dashed line in Fig. 4A). This high efficiency in small geometries thus suggests that the nasal cavities of small animals have simpler cross-sections, which has, indeed, been observed (2, 3).

Strikingly, one point that deviates strongly from the theoretical prediction in Fig. 4A is for humans. The surface area of their nasal cavity is about half of what the scaling suggests, and the volume is even only about 10% of the prediction (Fig. 4A, Inset). The data point for humans was calculated from a standardized nasal cavity, which was obtained by averaging reconstructed geometries of 30 humans (23), together with typical respiratory parameters given in Table 2. To examine the geometry of human nasal cavities more closely, we compute the local gap width in the standardized nasal cavity using the shape diameter function, which gives the average distance of nearby walls at every point of the surface (24) (Fig. 4B), and the hydraulic radius Rh=2A/Γ from the cross-sectional area A and the perimeter Γ (Fig. 4C). Both quantifications indicate that the gap width is remarkably constant over a large fraction of the standardized nasal cavity, while the cross-sectional area varies significantly (Fig. 4C, Inset). However, the measured Rh is significantly smaller than the predicted optimal gap width ℓ≈3.7mm, which follows from Eq. 7 together with the typical respiratory parameters summarized in Table 2. Thus, whereas the geometry of the human nasal cavity agrees with our qualitative result that the gap width should be constant for an efficient scalar exchange, the quantitative prediction deviate significantly from our theory. This is surprising since our theory worked well for all other tested mammals, and this might thus hint at an exceptional behavior of the human nasal cavity. Before we come back to this point in Discussion, we next consider how the shape of the head constrains the nasal cavity.

Geometric Constraints Imply Labyrinth-Like Cross-Sections.

Natural nasal cavities have a complex labyrinth-like cross-section, which does not resemble the theoretically optimal shape determined above. This difference is likely a consequence of other factors, including additional geometric constraints, since the wide shapes that we predict would simply not fit into the head. However, the fact that our theory agrees well with experimental data suggests that natural nasal cavities function close to optimally. This would suggest that the bending and branching of the nasal cavity that is necessary to obtain labyrinth-like geometries does not significantly affect the physical principles that led to the optimal gap width given in Eq. 7. To test this hypothesis, we examine the bending and branching of the cross-section and calculate how it affects the resistance and exchange efficiency. Fig. 5 shows that a U-shaped cross-section has virtually identical properties to a rectangle of the same aspect ratio. Consequently, bending the optimal cross-sectional shape in-plane affects neither the resistance nor the exchange efficiency significantly. To examine the consequence of branching, we consider a T-shaped junction with three rectangular branches of equal length. Numerical simulations indicate that both K and E are affected more strongly than in the case of bending, but still only change by a few percent compared with an equivalent rectangular shape; see Fig. 5. While the T-shaped junctions can be directly compared with rectangles, they also introduce additional corners, which increase the resistance. We show in SI Appendix that optimal junctions are more rounded and impact the resistance less than shown in Fig. 5. Taken together, neither bending nor branching affects the function of the nasal cavity strongly, implying that natural shapes are close to optimal.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Bending and branching does not affect the resistance K and scalar exchange efficiency E significantly. Shown are the prefactors CK (orange) and CE (blue) of (Left) bent and (Right) branched shapes normalized to the respective values for the corresponding rectangular shape with sides a and b as a function of its aspect ratio a/b. CK follows from a numerical solution of Eq. 1, while CE has been calculated from Eq. 4 using the first 16 eigenmodes of Eq. 5 as described in SI Appendix. Parameters are given in Table 2, and the dashed blue line in Right shows CE for channels that are elongated by a factor of 100, emphasizing the influence of the junction in longer channels.

Another geometric constraint on the nasal cavity comes from the fact that it must connect the pharynx (and thus the lungs) to the outside world. In humans, this forces a curved flow, which could influence the functions of the nasal cavity (5); see Fig. 1A. In general, curved flow increases the resistance and the exchange efficiency significantly (25, 26), but, in the case of the human nose, the bends are localized to the connecting regions, while the main nasal cavity is rather straight. We show in SI Appendix that the overall function of the nose is only slightly affected by the bent geometry, consistent with numerical simulations (27). This is because the connecting regions are much wider than the main nasal cavity. Note that this effect is even weaker in animals that have a straighter airflow than humans.

Gradients in the Scalar Exchange Limit Heat and Humidity Loss.

Up until now, we have derived the optimal geometry of the nasal cavity by focusing on the efficiency of heating and humidifying the inhaled air. However, improving this efficiency can come at the expense of heat and water loss during exhalation. This is because heating implies that the walls of the nasal cavity are warmer than the inhaled air, while the recapture of heat can only occur when the walls are colder than the exhaled air. Consequently, it is impossible to both heat the air efficiently and recapture most of the heat during exhalation. Such a conflicting requirement also holds for humidification, and we show in SI Appendix that small animals would lose about 1% of their body weight due to exhaled water each day. To understand how to prevent this loss while still heating and humidifying inhaled air, the scalar exchange needs to be studied for both inhalation and exhalation.

To study the trade-off between heating the inhaled air and recapturing heat during exhalation, we vary the scalar value cw that is prescribed at the walls of the nasal cavity. Given that cw is the same for inhalation and exhalation, we can calculate how the scalar transported with the air changes during these two processes. In particular, we can define a scalar exchange efficiency Eex for exhalation analogously to Ein for inhalation, given by Eq. 4. In our calculations above, we considered cw=cb and Ein=1, which implies Eex=0. Heat can be recaptured when cw is lowered to cw=0.5(ca+cb), where ca is the ambient value. A simple analytical model presented in SI Appendix shows that, in this case, Ein=Eex≈0.4. These values can be improved to Ein=Eex≈0.5 when a linear gradient from ca at the tip of the nose to cb at the nasopharynx is used. Numerical simulations presented in SI Appendix confirm this picture and show that the scalar exchange efficiency is actually higher than predicted by Eq. 6, because the scalar profile is typically not fully developed and entrance effects matter. Taken together, we conclude that a gradient boundary condition, as observed in nature (28), can improve the recapturing of heat and humidity, at the expense of a lowered exchange efficiency during inhalation. Note that this lower efficiency is approximately compensated by entrance effects that improve the exchange efficiency, so we expect Eq. 7 to work for nasal cavities with gradient boundary conditions and realistic lengths.