Lung diffusing capacity for nitric oxide and carbon monoxide following mild‐to‐severe COVID‐19

Abstract A decreased lung diffusing capacity for carbon monoxide (DLCO) has been reported in a variable proportion of subjects over the first 3 months of recovery from severe coronavirus disease 2019 (COVID‐19). In this study, we investigated whether measurement of lung diffusing capacity for nitric oxide (DLNO) offers additional insights on the presence and mechanisms of gas transport abnormalities. In 94 subjects, recovering from mild‐to‐severe COVID‐19 pneumonia, we measured DLNO and DLCO between 10 and 266 days after each patient was tested negative for severe acute respiratory syndrome coronavirus 2. In 38 subjects, a chest computed tomography (CT) was available for semiquantitative analysis at six axial levels and automatic quantitative analysis of entire lungs. DLNO was abnormal in 57% of subjects, independent of time of lung function testing and severity of COVID‐19, whereas standard DLCO was reduced in only 20% and mostly within the first 3 months. These differences were not associated with changes of simultaneous DLNO/DLCO ratio, while DLCO/VA and DLNO/VA were within normal range or slightly decreased. DLCO but not DLNO positively correlated with recovery time and DLCO was within the normal range in about 90% of cases after 3 months, while DLNO was reduced in more than half of subjects. Both DLNO and DLCO inversely correlated with persisting CT ground glass opacities and mean lung attenuation, but these were more frequently associated with DLNO than DLCO decrease. These data show that an impairment of DLNO exceeding standard DLCO may be present during the recovery from COVID‐19, possibly due to loss of alveolar units with alveolar membrane damage, but relatively preserved capillary volume. Alterations of gas transport may be present even in subjects who had mild COVID‐19 pneumonia and no or minimal persisting CT abnormalities. Trial registry ClinicalTrials.gov PRS: No.: NCT04610554 Unique Protocol ID: SARS‐CoV‐2_DLNO 2020.

urement of lung diffusing capacity for nitric oxide (DL NO ) offers additional insights on the presence and mechanisms of gas transport abnormalities. In 94 subjects, recovering from mild-to-severe COVID-19 pneumonia, we measured DL NO and DL CO between 10 and 266 days after each patient was tested negative for severe acute respiratory syndrome coronavirus 2. In 38 subjects, a chest computed tomography (CT) was available for semiquantitative analysis at six axial levels and automatic quantitative analysis of entire lungs. DL NO was abnormal in 57% of subjects, independent of time of lung function testing and severity of COVID-19, whereas standard DL CO was reduced in only 20% and mostly within the first 3 months. These differences were not associated with changes of simultaneous DL NO /DL CO ratio, while DL CO /V A and DL NO /V A were within normal range or slightly decreased. DL CO but not DL NO positively correlated with recovery time and DL CO was within the normal range in about 90% of cases after 3 months, while DL NO was reduced in more than half of subjects.
Both DL NO and DL CO inversely correlated with persisting CT ground glass opacities and mean lung attenuation, but these were more frequently associated with DL NO than DL CO decrease. These data show that an impairment of DL NO exceeding standard DL CO may be present during the recovery from COVID-19, possibly due to loss of alveolar units with alveolar membrane damage, but relatively preserved capillary volume. Alterations of gas transport may be present even in subjects who had mild COVID-19 pneumonia and no or minimal persisting CT abnormalities. Trial registry: ClinicalTrials.gov PRS: No.: NCT04610554 Unique Protocol ID: SARS-CoV-2_DLNO 2020.

K E Y W O R D S
alveolar membrane diffusive conductance, carbon monoxide, COVID-19, ground glass opacities, lung diffusing capacity, nitric oxide 1 | INTRODUCTION Infection with the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has been the cause, in a variable number of subjects, of a disease named severe coronavirus disease 2019  showing clinical manifestations ranging from mild upper airway symptoms to interstitial pneumonia with or without acute hypoxemic respiratory failure (Guan et al., 2020). Among the distinctive features of COVID-19, in comparison with influenza virus pneumonia, are an increase of serum Ddimer and, at autopsy, the presence of alveolar damage with widespread thrombotic microangiopathy (Ackermann et al., 2020). SARS-CoV-2 targets preferentially type II alveolar cells (Mason et al., 2020), which are the precursors for type I cells; thus, it can be hypothesized that COVID-19 survivors might develop gas exchange abnormalities because of aberrant alveolar wound healing, or loss of pulmonary vascular bed, or both.
Three preliminary studies found a mild decrement of lung diffusing capacity for carbon monoxide (DL CO ) in about half of subjects 1 month after symptom onset (Frija-Masson et al., 2020;Mo et al., 2020) or hospital discharge (Huang et al., 2020). Two studies found DL CO be reduced in 21% (Sonnweber et al., 2020) and 24% (Lerum et al., 2020) of subjects about 3 months after hospital admission, and one study in 34% of subjects 3 months after recovery from the acute phase of disease (van den Borst et al., 2020). Two of these studies (Lerum et al., 2020;Mo et al., 2020) also reported values of DL CO -to-alveolar volume (DL CO /V A ) ratio, that is, K CO , to be slightly decreased (Mo et al., 2020) or within the normal range (Lerum et al., 2020) in the majority of subjects, which would suggest an alveolar damage associated with diffuse microvascular destruction (Hughes & Pride, 2012). However, the interpretation of the above findings is complicated by differences in the cutoffs for defining DL CO abnormality, coexisting comorbidities, time of lung function studies, and severity of disease in the acute phase. Moreover, the major limit to lung CO uptake is its slow binding with intracapillary hemoglobin (Hb), which makes DL CO unable to distinguish between reductions of alveolar membrane diffusive conductance (DM) and pulmonary capillary blood volume (V C ) (Borland & Hughes, 2020;Guénard et al., 1987). By contrast, nitric oxide (NO) has a much greater affinity and faster reaction rate with Hb than CO (Gibson & Roughton, 1957), which make the lung diffusing capacity for NO (DL NO ) more sensitive to changes in DM than V C (Borland & Hughes, 2020;Guénard et al., 1987). Indeed, recent studies on interstitial lung diseases (Barisione et al., 2016(Barisione et al., , 2019 have shown that DL NO reflects fibrotic changes more accurately than standard DL CO . Thus, considering the complex pathophysiology of COVID-19 (Ackermann et al., 2020;Mason et al., 2020), we undertook the present study to investigate whether measurements of DL NO and DL CO can provide different information on gas exchange abnormalities persisting after COVID-19 that may be related to radiological findings, severity of pneumonia, and time of recovery.

| Study subjects
This study included 94 Caucasian subjects who attended our pulmonary function laboratory as outpatients for follow-up after in-hospital treatment for COVID-19 pneumonia, confirmed by ground glass opacities (GGO) or band-like consolidations on chest roentgenogram or computed tomography (CT) and positive nasopharyngeal swabs for SARS-CoV-2. Pulmonary function tests were obtained between 10 and 266 days after hospital discharge, which occurred only after each patient had been tested negative for SARS-CoV-2. To be included in the study, subjects were required not to have history of comorbidities potentially affecting lung diffusing capacity, that is, bronchial asthma, chronic obstructive pulmonary disease, pulmonary interstitial fibrosis or vasculitis, systemic collagen disease, congestive heart failure, liver or renal diseases, and morbid obesity. They were classified in three groups based on the presence or severity of acute hypoxemic respiratory failure and the respiratory support received during hospitalization (Table 1). Acute hypoxemic respiratory failure was diagnosed whenever the measured oxygen partial pressure (PaO 2 ) in an arterial blood sample drawn from the radial artery during room air breathing was below the age-adjusted lower limit of normal (Cerveri et al., 1995). The first group included 34 subjects who had no arterial hypoxemia, a second group included 34 subjects who had mild-to-moderate arterial hypoxemia treated by O 2supplementation with (n = 31) or without (n = 3) helmet continuous positive airway pressure, and a third group included 26 subjects who had severe arterial hypoxemia treated by O 2 -supplementation and invasive mechanical ventilation via tracheal intubation (n = 23) or tracheostomy (n = 3). During hospitalization, they had received antibiotics (n = 63), oral hydroxychloroquine (n = 49), corticosteroids (n = 43), enoxaparin (n = 37), tocilizumab or anakinra, (n = 24), and various antiviral drugs (n = 18). As a control group, we selected 31 healthy subjects, matched for anthropometric characteristics and smoking habit, among health professionals and their relatives studied before the onset of COVID-19 pandemic.

| Lung function measurements
Spirometry (Graham et al., 2019) and lung volumes (Wanger et al., 2005) were determined with subjects sitting in a | 3 of 10 BARISIONE ANd BRUSASCO whole-body plethysmograph (V62 J, SensorMedics-Viasys, CareFusion; Höchberg, Germany) and breathing quietly with a nose clip in place. Forced vital capacity (FVC), forced expiratory volume in one second (FEV 1 ), their ratio (FEV 1 /FVC), and total lung capacity (TLC) were measured and compared with predicted values (Quanjer et al., 1993(Quanjer et al., , 2012. Standard DL CO was measured (MasterScreen PFT System, Jaeger-Viasys, CareFusion, Höchberg, Germany) by single-breath technique with a measured breath-hold time of 11 ± 0.4 s. Maneuvers with inspired volume ≥85% of vital capacity, 8-12 s breath-hold time, and sample collection ≤4 s were retained for analysis . Results were compared with the predicted values from Stanojevic et al. (2017) after adjustment for effective Hb measured in available arterial or venous blood samples ( Hb meas (Cotes et al., 1972).
At least 5-10 min after standard DL CO , single-breath DL NO and DL CO were simultaneously measured with an actual breath-hold time of 5 ± 0.3 s as detailed elsewhere (Barisione et al., 2016(Barisione et al., , 2019, and the DL NO /DL CO ratio calculated. Predicted values for DL NO and DL NO /V A were from Zavorsky et al. (2017).
Personnel wearing equipment against exposure to SARS-CoV-2 did all testing and instrument cleaning disinfection procedures.
T A B L E 1 Subjects' anthropometric characteristics and lung function data (n = 125)

| Chest CT
In 38 subjects, a thin-section CT scan obtained between 0 and 207 days after hospital discharge and 34 days (median 8 days; interquartile range 25-75% [IQR 25-75% ] 0-19) before or after pulmonary function measurements was available. Scans of the entire chest were obtained in a supine position, during breath-holding at full inspiration, by a multi-detector row-spiral scanner (SOMATOM Emotion 6, Siemens AG Medical, Forchheim, Germany). Images were acquired by 110 kVp tube voltage at 1.25-mm slice thickness and reconstructed at 1-mm increments using smooth (B41 s) and sharp (B70 s) convolution kernels. CT scans acquired at an absolute lung volume ≥80% of plethysmographic TLC were retained for semiquantitative calculation of voxel percentages with GGO at six axial levels (Barisione et al., 2016(Barisione et al., , 2019 and automatic quantitative 3D analysis of mean lung attenuation (MLA) and its coefficient of variation (MLA CV %) for the entire lung (ITK-Snap 3.8.0, Philadelphia, PA, US) (Yushkevich et al., 2006).

| Statistical analysis
For each lung function measure, we calculated the percentage of predicted and z-score values. As lower limits of normality for DL NO and standard DL CO , we considered both 5th (LLN 5 , z-score −1.645) and 2.5th (LLN 2.5 , z-score −1.96) percentiles of the reference population. Categorical variables were compared by z-test with Yates correction, while Fisher's exact test was used to compare their distributions. Continuous variables were tested by one-way pairwise ANOVA with Holm-Sidak post hoc test for multiple comparisons. Associations between variables were tested for significance by the coefficient of determination (R 2 ). The difference between two dependent correlations with one variable in common was calculated by an asymptotic two-tailed z-test, with values >1.96 considered significant (Steiger 1980). Data are presented as mean ±SD or median with IQR 25-75% whenever appropriate. In all analyses, the acceptable type I error was set at p < 0.05.

| RESULTS
Collectively, all standard lung function measures and DL NO were significantly lower in the three COVID-19 groups than in the control group, whereas DL NO /V A and DL NO /DL CO ratios did not differ significantly.
There was a significant correlation between DL NO and standard DL CO z-scores (R 2 : 0.59; p < 0.0001) (Figure 1a). However, considering individual data, 35 subjects (37%) had DL NO but not DL CO below the LLN 5 , and 30 of them also below the LLN 2.5 , 19 subjects (20%) had both DL NO and DL CO below the LLN 5 and 16 of them also below the LLN 2.5 , 40 subjects (43%) had both DL NO and DL CO above the LLN 5 and 47 of them also above the LLN 2.5 , and only one subject had DL CO but not DL NO below the LLN 2.5 . There were no significant differences in the distribution of subjects with reduced DL NO , DL CO , or both in relation to the presence or severity of acute hypoxemic respiratory failure and type of respiratory support received during hospitalization. The DL NO /DL CO ratio was in the majority of COVID-19 subjects within 1.96 SD of the values observed in the control group (Figure 1b). F I G U R E 1 Panel a: Relationship between z-scores of standard lung diffusing capacity for carbon monoxide (DL CO ) and lung diffusing capacity for nitric oxide (DL NO ). Horizontal and vertical lines correspond to the 5th (dashed) and 2.5th (dotted) percentiles of reference values, that is, −1.645 and −1.96 z-scores, respectively. The numbers within brackets indicate the subjects falling into each quadrant (Q 1 -Q 4 ) bounded within 5th or 2.5th percentiles. Symbols indicate subjects recovering from mild (white), moderate (gray), and severe (black) COVID-19 pneumonia. Panel b: Correlation between simultaneous measures of DL NO and DL CO . Upper and lower oblique dashed lines indicate the 95% confidence interval for DL NO /DL CO ratio in healthy controls | 5 of 10

BARISIONE ANd BRUSASCO
There was a weak albeit significant positive correlation between standard DL CO (R 2 = 0.06; p = 0.014) but not DL NO (R 2 = 0.02; p = 0.15) and the time elapsed between negative test for SARS-CoV-2 and lung function studies (Figure 2a,b). Notably, of the 58 subjects studied after 3 months, 30 had DL NO below the LLN 5 and 25 also below the LLN 2.5 , while only six had DL CO below the LLN 5 and LLN 2.5 (p < 0.001).
The CT scans obtained within 34 days from lung function studies showed GGO above 5% of total lung volume be present in 21 (55%) of the 38 subjects examined (Figure  3a,b). Both DL NO and standard DL CO z-scores were inversely related to the extent of GGO with correlation coefficients insignificantly different (p = 0.61) between each other but y-intercepts significantly (p < 0.0001) lower for DL NO than standard DL CO . Therefore, reduced DL NO was associated with GGO more frequently than DL CO . Similar correlations were observed between DL NO or standard DL CO with MLA or MLA CV% (Figure 3c-f). Figure 4 shows an example of wide discrepancy between DL NO and standard DL CO in a subject with moderate CT abnormality. Quantitative analysis of the entire lung and qualitative analysis at six axial levels did not reveal areas of reticular opacities, honeycombing, or hypoattenuation (<−950 HU) in any subject.

| DISCUSSION
The main findings of the present study are that 1) abnormal DL NO was present in more than half of the subjects over 8 months of recovery from mild-to-severe COVID-19 pneumonia, whereas standard DL CO was abnormal in only 20%, 2) standard DL CO but not DL NO was positively correlated with recovery time, and 3) both standard DL CO and DL NO were inversely correlated with persisting CT abnormalities, but DL NO was more frequently associated with their presence.

| Comments on methodology
In this study, we measured DL CO by standard technique and in combination with DL NO , which required breath-hold times of 11 ± 0.4 s and 5 ± 0.3 s, respectively. Such a difference seems to have a negligible effect on final values of DL CO both in healthy subjects and restrictive disorders, that is, idiopathic pulmonary fibrosis (Barisione et al., 2016) and systemic sclerosis-associated interstitial lung disease (Barisione et al., 2019). Also in the present investigation, absolute values of DL CO measured by the two methods were strongly correlated (R 2 = 0.85; p < 0.0001) (Figure 5a) without systematic differences (Figure 5b). Therefore, we used standard DL CO values for comparison with DL NO and the results of previous studies.
Although the 5th percentile (z-score −1.645) is generally assumed as the lower limit of normal for standard lung function measurements including DL CO (Quanjer et al., 1993), the 2.5th percentile (z-score −1.96) has been suggested for DL NO with the currently available predictive equations (Munkholm et al., 2018;Zavorsky et al., 2017). Therefore, we have used both LLN 5 and LLN 2.5 to reduce false negative or false positive biases. As reference values for DL NO and DL NO /V A , we used the set of equations that provided the lower SD of zscores from our local data set of healthy subjects, that is, 0.71 and 0.70, respectively.
The alveolar concentration of endogenous NO increases in several inflammatory interstitial lung diseases (Cameli et al., 2020), which could theoretically bias DL NO measures. However, the mean NO concentration in the gas mixtures inhaled in the present study was 63.7 ± 10 ppm, resulting in alveolar concentrations ranging from 5.4 to 21.9 ppm, thus >1,000 times the threshold considered as a marker of pulmonary alveolitis. Hence, it is reasonable to assume that any effect of endogenous NO backpressure on DL NO measurements was negligible. Furthermore, 40 ppm of NO in the inspired gas could decrease hypoxic pulmonary vasoconstriction F I G U R E 2 Relationships between standard DL CO (panel a) or DL NO (panel b) and time elapsed from negative testing for SARS-CoV-2 to lung function studies. Symbols indicate subjects who recovered from mild (white), moderate (gray), and severe (black) COVID-19 pneumonia. Horizontal lines correspond to the 5th (dashed) and 2.5th (dotted) percentiles of reference values, that is, −1.645 and −1.96 z-scores, respectively. The shaded areas include the subjects with abnormal standard DL CO or DL NO values after the first 3 months of recovery Days after tested negative for SARS-CoV-2 (a) (b) (Glenny & Robertson, 2011), but this effect was observed withPAO 2 <60 mmHg (Asadi et al., 2015), thus well below the 102 ± 4 mmHg of this study.
The present study has two major limitations. First, lung function tests were obtained in a sitting posture and CT in supine position. The latter might have increased V C (Cotton et al., 1990), thus possibly affecting differently the relationships of DL NO and standard DL CO with CT density distribution data. Second, the study was cross-sectional, which may limit the clinical relevance of results but does not seem to invalidate their pathophysiological meaning and interpretation.

| Comments on results
To our knowledge, this is the first study using DL NO and DL CO to investigate the pathophysiology of alveolar-to-capillary gas exchange in patients recovering from COVID-19.

F I G U R E 3
Correlations between standard DL CO (panels a, c, and e), or DL NO (panels b, d, and f) and ground glass opacities (GGO), as percentage of total CT volume, mean lung attenuation (MLA) in Hounsfield units (HU), and its coefficient of variation (MLA CV%). Symbols indicate subjects who recovered from mild (white), moderate (gray), and severe (black) COVID-19 pneumonia. Horizontal lines correspond to the 5th (dashed) and 2.5th (dotted) percentiles of reference values, that is, −1.645 and −1. Clinically, COVID-19 pneumonia is associated in a variable number of subjects with acute hypoxemic respiratory failure ranging from mild-to-severe, whereas other subjects have no apparent gas exchange abnormalities (Guan et al., 2020). At autopsy of patients who died from severe COVID-19, diffuse alveolar damage, capillary endothelialitis, and fibrinous microthrombi with angiogenesis within the interalveolar septa has been observed (Ackermann et al., 2020). A question is whether these abnormalities occurring in the acute phase of the disease might leave late pathophysiological sequelae over the recovering phase and these depend on the presence or severity of acute hypoxemic respiratory failure. A mild reduction of standard DL CO has been reported in about half of survivors as early as 30 days after acute infection (Frija-Masson et al., 2020;Mo et al., 2020) or hospital discharge (Huang et al., 2020). In the present study, we found a much lower prevalence of decreased standard DL CO , that is, 20% and 18% with LLN 5 and LLN 2.5 , respectively, over 8 months after negative SARS-CoV-2 testing. There are three main reasons that may have contributed to this discrepancy. First, we used lower limits of normal based on z-scores instead of 80% of predicted (Huang et al., 2020;Mo et al., 2020), which tends to overestimate the presence of abnormality due to age-, sex-, and size biases (Miller & Brusasco, 2016). Indeed, our results are in keeping with the decrease of DL CO found in 24% of subjects in one study using z-scores (Lerum et al., 2020). Second, the proportion of subjects with reduced DL CO tended to decrease with the time elapsed from the negative testing for SARS-CoV-2 as suggested by Sonnweber et al. (2020). Instead, we found that more than half of subjects had DL NO below the LLN 5 and 49% below the LLN 2.5 , and this proportion remained near constant over 8 months. Third, almost all previous studies included several patients with comorbidities potentially affecting the final value of DL CO independent of COVID-19 severity (van den Borst et al., 2020;Frija-Masson et al., 2020;Huang et al., 2020;Mo et al., 2020;Sonnweber et al., 2020). Collectively, our results support the hypothesis that a more severe and prolonged abnormality of DL NO may be present after COVID-19 pneumonia, reflecting a prevailing decrement of DM. Several physiological mechanisms can explain a disproportionate reduction of DL NO and DL CO . Since the alveolarto-capillary transfer of CO is mostly limited by its slow reaction rate with Hb (Carlsen & Comroe, 1958), DL CO is relatively less sensitive to changes in DM than V C . By contrast, NO has a much greater affinity and fast reaction rate F I G U R E 4 Axial CT scan acquired at the bifurcation of main bronchi (carina) in supine position in a representative subject who had severe COVID-19 pneumonia treated by invasive mechanical ventilation. Note the discrepancy between DL NO and standard DL CO in the presence of moderate GGO extent. Abbreviations as in Table 1 Absolute % predicted z-score with Hb, which make DL NO more sensitive to DM than V C (Borland & Hughes, 2020). Thus, the findings of the present study suggest that a decreased DM is more frequent and persistent than the reduction of V C in the recovery phase after COVID-19 pneumonia. One reason for decreased DM could be simply a loss of lung volume, but this would have caused an increase of DL NO /V A , which was instead slightly below the LLN 5 or LLN 2.5 in about one third of subjects. Moreover, TLC was significantly lower than in controls and in subjects with moderate-to-severe than mild pneumonia, while there were no differences in the distribution of DL NO and DL CO abnormalities. DL NO /DL CO ratio was in most cases within the normal range suggesting that alveolar damage rather than loss of lung volume was the major determinant of diffusion limitation (Hughes & van der Lee, 2013). A possible mechanism for the differences in DL NO and DL CO over recovery time could be that SARS-CoV-2, by targeting type II and eventually type I pneumocytes (Mossel et al., 2008), may cause a persistent damage of alveolar membrane while vasculopathy with capillary microthrombi is possibly reversing more rapidly after the acute phase of the disease. However, while V C reflects pulmonary blood volume only, DM reflects alveolar membrane thickness and surface but also vessel surface (Kang & Sapoval, 2016). The latter may be reduced as a consequence of capillary remodeling or obliteration with blood volume being redistributed to unaffected lung regions (Oppenheimer et al., 2006;Pande et al., 1975), or uneven red cell distribution within the alveolar capillaries (Hsia et al., 1997). Another reason for decreased DM without V C changes could be the presence of interstitial edema (Zavorsky et al., 2014), which would be consistent with the closer associations of DL NO than standard DL CO with CT abnormalities. In the present study, GGO was the only qualitative CT abnormality persisting after COVID-19 and correlated with decrement of DL NO and standard DL CO . In interstitial pulmonary fibrosis (Barisione et al., 2016) or interstitial lung disease associated with systemic sclerosis (Barisione et al., 2019), we found DL NO be correlated with CT fibrotic abnormalities but not GGO. This may suggest that interstitial edema by itself may not be sufficient to alter substantially the alveolar-to-capillary gas transport, owing to the high solubility of both NO and CO (Wilhelm et al., 1977). Moreover, we observed reduced DL NO even in the absence or minimal-to-moderate GGO, which suggests that mechanisms other than alveolar membrane thickening may contribute to diffusion abnormality after COVID-19.

| CONCLUSIONS
In subjects recovering from COVID-19 pneumonia, DL NO is impaired more frequently and more persistently than standard DL CO , suggesting an impairment of DM due to alveolarcapillary damage and loss of alveolar units with V C relatively preserved. DL NO was more frequently abnormal than standard DL CO even in subjects with minimal or absent CT abnormalities, suggesting persistent alveolar damage in these subjects. Further long-term studies are necessary to investigate whether these medium-term changes may evolve into chronic morphological and functional abnormalities.

CONFLICT OF INTEREST
G.B. and V.B have no financial/nonfinancial interests to disclose.

ETHICAL APPROVAL
The study was approved by the Regional Ethics Committee (CER Liguria Registry No.: 412/2020 -DB id 10794) and each subject gave written informed consent to use his/her anonymized personal data.