Human Subjects

The University of California, San Diego Human Research Protection Program reviewed and approved this study. Six subjects, the same from Geier et al. 2018 (8), were re-recruited, each passed a standard MRI safety screen, and all subjects denied any history of cardiopulmonary disease. Each subject provided written, informed consent to participate in the study. Forced expiratory volume in one second (FEV₁) measurements were made using an Easyone diagnostic Spirometer (ndd Medical Technologies, Inc., Andover, MA), and percent predicted values were obtained by comparison to National Health and Nutrition Examination Survey III reference values (9). Subjects had baseline FEV₁ values of 101% (6%). Two additional imaging sessions (prone methacholine/supine imaging and prone methacholine/prone imaging) were acquired; the subjects’ concentration of methacholine that elicits 20% drop in FEV₁ (PC₂₀) values [6.4 mg/ml (SD 3.4)], methacholine doses [9 mg/ml (SD 5)] and imaging data from the supine baseline and first methacholine challenge (supine methacholine/supine imaging) from the previously published study were reused (8). Subject data as they were acquired for the previous study (8) are shown in Table 1. The time between study sessions, for each subject, is shown in Table 2. Although all subjects completed all sessions of the study, a technical issue discovered in postprocessing necessitated that the prone baseline and prone methacholine/prone imaging data for two subjects (S2 & S4) be excluded from the analysis (see Experimental setup below).

Methacholine Bronchoconstriction

In our previous study (8), each subject underwent a standard methacholine challenge test to determine the appropriate concentration of drug that could be used to elicit a moderate bronchoconstriction in the subsequent imaging sessions. Methacholine challenge testing was performed according to the standard procedures established by the American Thoracic Society (24) with the sole modification that the subject was supine during all parts of the challenge. Provocholine (methacholine chloride USP, Methapharm, Inc., Brantford, Canada) was nebulized and administered via a Koko Dosimeter (nSpire Health, Inc., Longmont, CO) in a stepwise fashion, starting at 0.03125 mg/ml and doubling each step until a 20% drop in FEV₁ was achieved. In these subjects, the maximum PC₂₀ dose was 16 mg/ml.

For each subject, the methacholine concentration chosen for the rest of the study (Table 1) was the final concentration administered in the stepwise challenge, i.e., the lowest that elicited PC₂₀. Methacholine was administered via a Koko Dosimeter before imaging in either the prone or supine posture, depending on the specific imaging session (Table 2).

Methacholine action peaks between 1 and 4 min after inhalation, reaches a steady state that lasts for an average of 75 min, and then fades for an average of 57 min (5). In all three visits, the specific ventilation (SV) imaging (SVI) protocol took place 15–35 min after methacholine administration in the steady-state period of methacholine action so that SV was as constant as possible during the 20-min acquisition. Methacholine was administered to all subjects in all sessions by the same individual. Each subject was given the same instructions so that breathing during drug inhalation was as standardized as possible between subjects.

Imaging Sessions

MR imaging was performed twice at baseline, once while supine and once while prone, and three times following bronchoconstriction by methacholine. The subjects’ posture was modified during methacholine administration and imaging as part of the experiment. The postural conditions were 1) supine methacholine and supine imaging, 2) prone methacholine and supine imaging, and 3) prone methacholine and prone imaging. In the second condition, the subjects were turned from prone to supine directly before imaging, ~10 min after methacholine inhalation. The delay between administration and repositioning was included to ensure that the subjects had reached, while prone and before repositioning, the steady state phase of methacholine action (5). The data for supine baseline and supine methacholine/supine imaging condition were acquired in a previous study (8). The supine methacholine/supine imaging condition is the same as that from the first constriction (day 1) in that previous study.

FEV₁ measurements were made before imaging for the baseline scans and between methacholine inhalation (~2 min postadministration) and imaging for the bronchoconstricted scans. Supine baseline imaging was performed the same day as condition 1, and prone baseline imaging was performed the same day as condition 3. Imaging sessions were always performed in order of 1–3. The time intervals between study sessions are shown in Table 2. The long duration of the study reflects the fact that subjects were re-recruited from the prior study, and day 1 of the previous study was treated as day 1 of the current study.

SVI

SVI is described in detail in Sá et al. (27). This implementation of SVI was the same as in our previous work (8), with the sole exception that the subjects’ posture during methacholine delivery and MR imaging was modified. A brief description of SVI is presented here.

Image processing.

Image processing was performed with custom-written Matlab (Mathworks, Natick, MA) code. Each scan was treated independently, and comparisons between conditions were performed after processing.

For each sagittal slice imaged (4 from each scan) a 220-image time series was created. A generalized dual-bootstrap iterative closest point algorithm (34), used in affine transformation mode, was applied to register images to a selected reference image. Images with a >10% change in imaged lung volume because of transformation were removed from the time series and treated as missing data, based on previous work that has shown that deformations up to 10% volume change can be accurately registered using an affine transformation (1).

Intensity time series were created for voxels within a manually drawn region of interest, and each voxel’s time course was correlated with 50 modeled time courses for SV values equally spaced in log₁₀ from 0.01 to 10. The modeled time course that shared the maximum correlation coefficient with the measured time course was selected as the voxel’s SV. The spatial map of all voxel SV values was smoothed using a geometric algorithm with a kernel size of 7 voxels (~1 cm²). Maps from the four sagittal slices in each scan were combined to create a histogram of voxel-SV values for the entire imaged right lung (Fig. 1).

Download figureDownload PowerPoint

SV Heterogeneity

SV heterogeneity, as has been reported by our group previously (26), was taken as the full-width at half maximum of a log(Gaussian) function fit to the SV histogram of the entire imaged right lung (Fig. 1). This metric was computed for the two baseline scans and three bronchoconstricted scans for each subject.

Identification of bronchoconstricted regions.

SV maps were normalized between conditions to account for changes in global SV, which can be the result of changes in tidal volume and/or FRC. Table 3 shows the prenormalized mean SV and imaged thoracic cage volumes for each condition.

Table 3. Global metrics of constriction for all six subjects in all imaging conditions

	FEV1, l/s (% Change From Baseline)
	Supine baseline†	Prone baseline	Supine Mch/supine imaging†	Prone Mch/supine imaging	Prone Mch/prone imaging
S1	3.55	3.80	2.86 (−19)	2.94 (−23)	3.40 (−11)
S2	2.91		2.53 (−13)	2.49 (−19)
S3	4.99	5.01	3.92 (−21)	4.11 (−18)	4.10 (−18)
S4	3.05		2.34 (−22)	2.84 (−16)
S5	3.70	3.56	3.14 (−15)	2.99 (−19)	2.54 (−29)
S6	3.11	2.95	2.38 (−22)	2.06 (−30)	2.59 (−12)
Mean ± SD	3.53 ± 0.78	3.83 ± 0.86	2.86 ± 0.60 (−19 ± 4)	2.91 ± 0.69 (−20 ± 6)	3.16 ± 0.74 (−17 ± 8)
Mean Specific Ventilation, Prenormalization (% Change from Baseline)
S1	0.25	0.10	0.10 (−61)	0.19 (−24)	0.12 (26)
S2	0.17		0.09 (−46)	0.08 (−53)
S3	0.26	0.13	0.28 (5)	0.23 (−12)	0.15 (16)
S4	0.16		0.13 (−16)	0.20 (23)
S5	0.33	0.24	0.33 (−2)	0.31 (−6)	0.14 (−43)
S6	0.37	0.13	0.21 (−42)	0.15 (−60)	0.16 (22)
Mean ± SD	0.26 ± 0.08	0.15 ± 0.07	0.19 ± 0.10 (−27 ± 27)	0.19 ± 0.08 (−22 ± 31)	0.14 ± 0.01 (5 ± 32)
Imaged Thoracic Cage Volume, ml (% Change from Baseline)
S1	851	844	927 (9)	831 (−2)	811 (−4)
S2	600		732 (22)	655 (9)
S3	930	980	1,206 (30)	1,141 (23)	1,371 (40)
S4	559		785 (40)	698 (25)
S5	509	736	518 (2)	556 (9)	827 (12)
S6	497	551	561 (13)	567 (14)	637 (16)
Mean ± SD	658 ± 186	778 ± 181	788 ± 254 (19 ± 14)	741 ± 220 (13 ± 10)	912 ± 318 (16 ± 18)
FWHM of SV Distribution (% Change from Baseline)
S1	0.44	0.47	0.92 (110)	0.79 (80)	0.43 (−9)
S2	0.32		1.03 (224)	0.79 (343)
S3	0.43	0.48	0.91 (111)	0.99 (129)	0.93 (93)
S4	0.57		0.46 (−20)	0.49 (−15)
S5	0.61	0.52	0.55 (−9)	1.12 (85)	0.92 (78)
S6	0.32	0.52	1.00 (210)	1.86 (475)	0.94 (80)
Mean ± SD	0.45 ± 0.12	0.50 ± 0.02	0.81 ± 0.25 (104 ± 104)	1.11 ± 0.48 (183 ± 186)	0.80 ± 0.25 (60 ± 47)
Volume Fraction of Lung Constricted
S1			0.32	0.27	0.06
S2			0.25	0.32
S3			0.38	0.34	0.39
S4			0.34	0.20
S5			0.19	0.19	0.29
S6			0.27	0.30	0.31
Mean ± SD			0.29 ± 0.07	0.27 ± 0.06	0.26 ± 0.14

The two left columns show the baseline (premethacholine) values of each metric measured in both the supine and prone postures. See text for details. Postmethacholine values are shown in the right three sets of columns with headings referring to their respective conditions. Relative changes from baseline are indicated as percentages. Only four of our six subjects had valid data for the last study, prone methacholine/prone imaging, because of an error in data acquisition (see methods). FEV₁, forced expiratory volume in one second; FWHM, full width half maximum; Mch, methacholine; SV, specific ventilation.

^†Indicates data from our previously published study (8).

Following normalization, bronchoconstricted maps were spatially correlated with their respective baseline maps, prone to prone and supine to supine. Because the average imaged lung volume during bronchoconstriction was 17% ( SD 13) larger than at baseline, a difference that exceeded the 10% limit to which we can confidently register using an affine transformation, we could not use our standard image registration approach. Instead, we divided each multislice set of images into 12 (dependent-nondependent) × 12 (apical-basal) × 4 (medial-lateral, defined by slice) volume elements, the extent of which were based on the size of the lung, redefined for each condition. Comparisons were then made between volume elements in different conditions with the same 12 × 12 × 4 coordinate (Fig. 2). Each of the 12 × 12 × 4 volume elements contained typically between 25 and 100 voxels (depending on its location in the lung). Individual voxels were not compared between conditions.

Download figureDownload PowerPoint

For the purposes of this study, a volume element i was considered constricted if

$${\text{SV}}_{\text{challenge}}^i<0.5*{\text{SV}}_{\text{baseline}}^i\hspace{0.17em}\text{and}\hspace{0.17em}{\text{SV}}_{\text{challenge}}^i<{\overline{\text{SV}}}_{\text{baseline}}$$(1)

where $\overline{\text{SV}}$ represented the center of the SV distribution.

This imposed two conditions for the constriction designation: 1) SV must have decreased significantly from baseline and 2) that a volume element labeled “constricted” was ventilating less than average. These criteria were designed to prevent two types of lung regions from being misclassified as constricted, 1) regions with consistently low ventilation (even at baseline) that were not necessarily responding to methacholine and 2) regions with unusually high SV at baseline that may have experienced a reduction in ventilation because of methacholine but not one severe enough to hamper gas exchange.

The constriction designation we use in this study is somewhat analogous to the ventilation defect designation that appears in other, similar studies (10–12, 21, 32, 33). Ventilation defects are clusters of voxels that represent ventilatory slow space. They can be identified through at least three different methods: manually by an experienced observer, through hierarchical k-means clustering (18), or by spatial fuzzy c-means clustering (15).The lung elements in our study that are designated as constricted certainly include voxels that would be designated as ventilation defect, but they also include others that strongly respond to methacholine, as indicated by their >50% drop in FEV₁ but not to the point that they would be considered “defects.” For example, a lung element with a relatively high SV (say of 0.3 compared with a lungwide mean SV of 0.2) that fell to an SV of 0.14 (compared with a lungwide mean SV of 0.2) would be considered constricted but would probably not be considered to be part of a defect. Conversely, an element with extremely low SV (say 0.02 compared with a lung-wide mean of 0.2) at baseline that stayed the same after methacholine may be considered a defect in both conditions but would not be classified as constricted in our analysis.

Height from the dependent lung as a predictor of constriction.

The vertical height of each 12 × 12 × 4 volume element from the dependent lung border was measured and, along with constriction outcome for each condition, concatenated for all subjects. For each postural condition, a logistic regression model was used to estimate the how the ln(odds) of constriction (our binary outcome variable) varied as a function of a volume element’s height (Fig. 3).

Download figureDownload PowerPoint

Statistics

R (25) was used for all statistical analyses performed in this study. Paired t-tests were performed using the stats package (25) to determine how prone versus supine posture affected baseline measurements of FEV₁, SV heterogeneity, mean SV, and thoracic cage volume. Only data from subjects with both measurements were used for this last computation.

Linear mixed effects models were generated using the lme4 package (2) to investigate the dependence of metrics of constriction (Table 2, Fig. 4) on the postural condition of the experiment, given the missing measurements for two subjects. For each model, the postural condition of the experiment (supine methacholine/supine imaging, prone methacholine, supine imaging, etc.) or pair of experiments was taken as a fixed effect and subject was taken as a random effect. P values were obtained by a likelihood ratio test of the model with the fixed effect (postural condition) versus a null model with no fixed effect.

Logistic regression models were fit using the glm package (25) to characterize how the likelihood of constriction varied as a function of height above the dependent lung. McFadden’s pseudo-R² was used to estimate the strength of the model fit (22). Formally, McFadden’s R² = 1 − ln(L_M)/ln(L₀) where L_M is the likelihood for the model with the chosen predictors and L₀ is the likelihood for a null model with no predictors. A higher value of McFadden’s R² means that more of the variance in the data is being explained by the logistic regression model (14). The P value (Wald Test) for each model and the fit coefficient for height are also reported in Fig. 3. To characterize the uncertainty of our logistic regression models, we used a data-driven bootstrapping approach. For each postural condition, we resampled with replacement from our measured data set of height versus constriction (keeping the links between the two intact) until a set of equal size to the original was obtained. A logistic regression was then fit to this resampled data set, and the P value, fit coefficients, and McFadden’s R² of the model were recorded. We repeated this process 1,000 times and took the standard deviation of each set of 1,000 outputs to be the uncertainty of our original fit.

A log transformation was applied to the odds ratios account for their intrinsic positive skew and thus make statistical testing possible (3, 4). To help interpret the odds ratios for repeat constriction, we modeled the null hypothesis that constriction in the first condition does not predict constriction in a subsequent condition. To do this, we computed the odds ratio for repeat constriction between the first condition, supine methacholine/supine imaging, and 1,000 modeled data sets in which all subjects had randomly allocated (but proportionally representative, as defined by each subject’s average fraction of constricted lung between the subsequent two conditions) constriction. The average of these 1,000 modeled study pairs was taken to represent the odds ratio for repeat constriction, assuming no spatial specificity for constriction in the second condition. A bootstrap randomization test with 1,000 iterations was used to determine if the observed ln(odds) for repeat constriction were statistically different from the null model for each pair of studies. Because four comparisons were performed (one for each pair of studies), the P values from these bootstrap tests were Bonferroni-corrected.

Similarly, we modeled a first-order hypothesis that constriction is predicted only by the effect of height above the dependent lung border as characterized by our best-fit logistic regression models (Fig. 3). For this, 1,000 modeled data sets for each postural condition were computed in which each lung element’s likelihood of constriction was determined by its height above the dependent lung. The odds ratio for constriction in the model data, given constriction in the observed data, was then computed for each of the 1,000 repeats for each comparison of interest. The mean and standard deviation of these 1,000 odds ratios were taken to be representative of the odds ratio for repeat constriction within/between postural conditions, given the predisposing factor of height above the dependent lung alone.

Throughout the text, results are presented as mean (standard deviation).

Baseline Measurements

At baseline neither FEV₁, nor the full width half maximum of the SV distribution, nor the thoracic cage volume varied between the prone and supine postures (P = 0.89, P = 0.47, and P = 0.21, respectively, two-tailed paired t-tests). Mean SV at baseline was lower while prone than while supine (P = 0.02).

Lung-Wide Metrics of Constriction

FEV₁ decreased for all subjects following methacholine (as imposed by the study design), and the heterogeneity of SV increased on average (Table 3). Linear mixed effects models showed that posture during methacholine inhalation and/or imaging had no significant fixed effect on the methacholine-induced fall in FEV1 (P = 0.96) and no significant fixed effect on the increase in SV heterogeneity as measured by the change in the full width half maximum of the subjects’ SV distributions (P = 0.09). The percent change in mean SV did not differ significantly based on the subjects’ posture during methacholine delivery and/or imaging (P = 0.18), nor did the change in thoracic cage volume (P = 0.17).

On average, 29% (SD 7) of the lung (by volume) constricted following supine methacholine inhalation when the subject was imaged supine, 27% (SD 6) of the lung constricted following prone methacholine inhalation when the subject was imaged supine, and 26% (SD 14) of the lung constricted following prone methacholine inhalation when the subject was imaged prone (Table 3). A linear mixed effects model showed no significant fixed effect of this metric on posture during methacholine and/or imaging (P = 0.80).

Height from the Dependent Lung Border as a Predictor of Constriction

Figure 3 shows the binary outcome variable, constriction/nonconstriction, as a function of height above the dependent lung border, along with the corresponding best fit logistic regression model. Bar graphs of the fraction of lung that constricted in each 1-cm bin of height are superimposed. A volume element’s height from the dependent lung border was a significant predictor of constriction in the two conditions in which imaging was performed supine (supine methacholine/supine imaging and prone methacholine/prone imaging) but not when imaging was performed prone (prone methacholine/prone imaging). The coefficient of the logistic regression was larger (more negative), and the predictive value of height (measured by McFadden’s pseudo-R²) was stronger in the prone methacholine/supine imaging condition than in the supine methacholine/supine imaging condition. The coefficient of a logistic regression, which is somewhat analogous to the slope of a linear regression, quantifies the rate by which the ln(odds) of an outcome (in this case, constriction vs. nonconstriction) changes as a function of the independent variable (in this case, height above the dependent lung).

Odds Ratio for Recurrent Constriction of a Volume Element

In our previously published study (8) we showed that a lung volume element that had previously constricted was 7.7 times more likely [odds ratio 7.7 (SD 3.5)] to constrict during a subsequent study in which posture was the same both during methacholine inhalation and while imaging. In this study, we computed the odds ratios for repeat constriction between data from the first imaging session of our previously published study, in which supine methacholine administration was followed by supine imaging, and the new data acquired during two altered experimental conditions, prone methacholine administration followed by supine imaging, and prone methacholine administration followed by prone imaging (Fig. 4). Furthermore, we computed the odds ratio for repeat constriction between the two new conditions in this study, prone methacholine/supine imaging and prone methacholine/prone imaging, to determine the effect of solely modifying posture during imaging. We found that the odds ratio for repeat constriction between supine methacholine/supine imaging and prone methacholine/supine imaging was 7.4 (SD 3.9), the odds ratio for repeat constriction between the supine methacholine/supine imaging condition and the prone methacholine/prone imaging condition was 0.8 (SD 0.6), and the odds ratio for repeat constriction between the prone methacholine/supine imaging and prone methacholine/prone imaging conditions was 1.2 (SD 0.9) (Fig. 4).

A linear mixed effects model showed that the log-transformed odds ratios for repeat constriction varied between study pairs (P < 0.001). One thousand iterations of a model of our experiment with random (but proportionally representative) patterns of constriction showed that completely random constriction in the second condition (for all subjects) would result in an average odds ratio of 1.0 (SD 0.2). For the two study pairs in which the posture during imaging was the same (supine), a previous constriction event in a lung element was a strong predictor for a subsequent constriction event; the ln(odds) of the data was greater than the than ln(odds) of the spatially random model, P = 0.004 for supine methacholine/supine imaging → supine methacholine/supine imaging and P = 0.006 for supine methacholine/supine imaging → prone methacholine/supine imaging (Bonferroni-corrected). For the two study pairs in which the posture during imaging was different (prone vs. supine) a previous constriction event in a lung element did not predict a subsequent constriction event; the ln(odds) of the data was not different than the than ln(odds) of the spatially random model, P > 0.99 for supine methacholine/supine imaging → prone methacholine/prone imaging, and P > 0.99 for prone methacholine/supine imaging → prone methacholine/prone imaging [Bonferroni-corrected]).

First-order models of constriction, which incorporated the logistic regressions depicted in Fig. 3, allowed us to calculate the expected odds ratio for recurrent constriction given the effect of height above the dependent lung. These models showed that the height dependence alone would result in an odds ratio of 1.3 (SD 0.4) for repeat constriction between two subsequent supine methacholine/supine imaging challenges, an odds ratio of 1.4 (SD 0.5) for repeat constriction between the supine methacholine/supine imaging and prone methacholine/supine imaging condition, and an odds ratio of 1.0 (SD 0.3) for repeat constriction between prone methacholine/supine imaging and prone methacholine/prone imaging condition.

Conclusion

The results of our study suggest that height above the dependent lung is a predictor of bronchoconstriction following methacholine when a subject lies supine but is not a predictor when a subject lies prone. This may suggest that differences in regional lung inflation play a role in determining the spatial pattern of methacholine bronchoconstriction, because lung inflation is more homogeneous while prone than while supine (13, 23). Furthermore, we have shown that changing posture during methacholine administration from supine to prone neither affects the spatial repeatability of constriction nor decreases the dependent (posterior) skew in constriction if imaging is subsequently performed supine. This suggests that the effects of the methacholine deposition pattern on the spatial pattern of bronchoconstriction, if they exist, can be overwhelmed by mechanical and/or anatomical factors.