Standardised tachypnoea—50% increase in the baseline respiratory rate—was successfully achieved in all preterm lambs under all the conditions (Supplementary Table S6). Apart from tachypnoea, close observation did not reveal any other clinical signs of respiratory distress in any animal.

On postnatal days 7–8, despite tachypnoea, none of the three NRS conditions negatively impacted the safety variables during bottle-feeding (Figure 1A and Supplementary Table S7). On the contrary, minimal HR and minimal SpO₂ were greater with nCPAP than the other conditions. None of the NRS conditions negatively impacted arterial blood gases and pH at baseline or 1 min after feeding (Supplementary Table S8). Analysis of videofluoroscopy did not reveal any laryngeal penetration or tracheal aspiration. Again, videofluoroscopic assessment often could not be performed due to the need for the experimenter’s assistance, especially with nCPAP (see Supplementary Table S1). As for SU–SW–BR coordination, SU–SU interval was greater with nCPAP than with HFNC; SU–SU COV was lower with HFNC than with the other three experimental conditions (Figure 2A and Supplementary Table S7). Lastly, efficiency of bottle-feeding was not altered by any NRS condition (Figure 3A and Supplementary Table S7).

On postnatal days 13–14, none of the three NRS conditions significantly impacted the safety and efficiency of bottle-feeding or the SU–SW–BR coordination (Figures 1B, 2B, 3B). Laryngeal penetration or tracheal aspiration was never observed when videofluoroscopic assessment could be conducted (see Supplementary Table S1). Compared to the control condition, SU–SU and SW–BR intervals were higher with nCPAP, and SU–SW interval was higher with HFNCcpap (Supplementary Table S9). In addition, apart from a slightly lower PaO₂ with HFNC and HFNCcpap compared to control condition at baseline, none of the NRS conditions negatively impacted blood gases before or 1 min after feeding (Supplementary Table S10).

Compared to experiments without tachypnoea, the SW–SW interval, as well as arterial CO₂ at 1 min after bottle-feeding was significantly higher on postnatal days 7–8 (Supplementary Tables S11, S12). In contrast, the following variables were significantly lower: baseline SpO₂, minimal SpO₂, SpO₂ decrease, arterial O₂ pressure both at baseline and 1 min after bottle-feeding, arterial pH at 1 min after feeding, as well as the SW–BR interval and SW–BR COV.

In addition, on postnatal days 13–14, mean SpO₂, minimal SpO₂, and arterial O₂ pressure were lower at baseline and 1 min after bottle-feeding (Supplementary Tables S11, S12).

In infants born prematurely, the maturation of SU–SW–BR coordination commonly allows for initiating oral feeding around 32–34 weeks post-gestational age (Mizuno and Ueda, 2003; Medhurst, 2005; Breton and Steinwender, 2008). At that time, however, the convalescent preterm infant often requires some form of NRS. As a result, caregivers are faced with the unresolved conundrum of risking either cardiorespiratory events during attempts at oral feeding under NRS or the consequences of delaying oral feeding. A recent survey of practice of NICUs and PICUs in Australia and New Zealand reported that most units do not feed on NRS as aspiration risks are unclear (Canning et al., 2020). In addition, a recent systematic review of clinical studies concluded that there is insufficient evidence to recommend the introduction of oral feeding whilst on nCPAP or HFNC to facilitate transition to full oral feeding without adverse effects (Canning et al., 2021).

Given the strong caregiver opinions, the lack of evidence-based guidelines, and the consequent high variability in feeding practices under NRS, we believe that carefully planned preclinical studies in an appropriate animal model can provide the physiological knowledge required to design and conduct safer and more decisive clinical studies in human infants. Our research programme in newborn lambs allows us to assess the safety and the SU–SW–BR coordination more thoroughly in the same subject under different NRS conditions—including no NRS—and under repeated radiation exposure. Our current study represents our unique attempt at mimicking a tachypnoeic preterm infant at ∼34 weeks of postconceptional age.

This study importantly revealed that oral feeding is safe under nCPAP at 6 cmH₂O, HFNC at 7 L•min^–1, or HFNCcpap in preterm lambs 7–8 days old, as well as after reaching full-term postconceptional age, even in the presence of tachypnoea. Indeed, none of the cardiorespiratory variables assessed were negatively impacted by any of the tested NRSs. This absence of any significant cardiorespiratory inhibition under all tested experimental conditions is a strong argument against laryngeal penetrations. Indeed, past studies conducted by our team in the same preterm lamb model have shown that laryngeal penetrations triggered foetal-type laryngeal chemoreflexes characterised by bradycardia, apnoea, oxygen haemoglobin desaturation, and laryngospasm (St-Hilaire et al., 2007). In addition to the absence of cardiorespiratory events, videofluoroscopy assessment, when it could be performed, did not reveal any laryngeal penetration or tracheal aspiration regardless of the experimental condition. These results agree with past studies reporting no clinical signs suggestive of aspiration in convalescing preterm infants with bronchopulmonary dysplasia or orally fed while on nCPAP (Hanin et al., 2015; Dalgleish et al., 2016; Jadcherla et al., 2016; Glackin et al., 2017; Shimizu et al., 2019). The results are, however, contrary to one single videofluoroscopy study showing a significant increase in deep laryngeal penetration and aspiration episodes in preterm infants bottle-fed on nCPAP (Ferrara et al., 2017). Of note, our recent observations in full-term lambs revealed very rare laryngeal penetrations and tracheal aspirations occurring almost exclusively with HFNC, especially with HFNCcpap (Alain et al., 2021). The reason for this slight discrepancy between our two studies is unclear. It could, however, be partly due to the impossibility of obtaining videofluoroscopic assessment under all experimental conditions in preterm lambs. Nevertheless, pooled results obtained in newborn lambs—full-term and preterm—clearly show that laryngeal penetration and tracheal aspiration remain rare occurrences, with no deleterious cardiorespiratory consequences.

We only found a few statistically significant differences supporting an impact of NRS on SU–SW–BR coordination. These differences, which seem to predominate on postnatal days 13–14, remain unexplained. The most notable results might be the increased SW–SW interval by all three NRSs in the absence of tachypnoea, as well as the increase in both SU–SU and SW–BR intervals under nCPAP in the presence of tachypnoea. These results are of uncertain physiological significance. Overall, we believe the two important messages are the following. First, none of the NRSs had a systematic effect on SU–SW–BR coordination. Second, whatever the NRS condition and the presence or not of tachypnoea, preterm lambs adopted an appropriate SU–SW–BR strategy to successfully feed orally without triggering deleterious cardiorespiratory reflexes.

One peculiar aspect of SU–SW–BR coordination in preterm subjects is worth discussing. It is well documented in both humans (Gewolb et al., 2001; Gewolb and Vice, 2006; Barlow, 2009) and animal models (Samson et al., 2017; Mayerl et al., 2019) that SU–SW–BR coordination is different in preterm and full-term infants. In particular, the percentage of nutritive swallows occurring during an apnoea is lower in full-term lambs (Samson et al., 2018) than in preterm ones (Samson et al., 2017), which is in agreement with results obtained in human infants (Lau et al., 2003; Mizuno and Ueda, 2003; Gewolb and Vice, 2006). This difference between preterm and full-term lambs, however, appears to hold true only when the bottle is held by an experimenter. Indeed, we recently observed that an important proportion—almost half—of nutritive swallows occurred during episodes of apnoea in full-term lambs also when they fed unassisted from our bottle-feeding device (Alain et al., 2021). This observation remains unexplained. The present results in preterm lambs unfortunately do not yield a clear picture of this aspect of SU–SW–BR coordination, for the latter might have been modified in the lamb needing assistance during multiple attempts at bottle-feeding.

In recordings obtained at days 7–8, no indices of bottle-feeding efficiency were modified by any of the NRSs tested. These results are in accordance with our most recent study in full-term lambs, which used a similar experimental design (Alain et al., 2021). The results herein, however, do not replicate those obtained previously in preterm and full-term lambs (Samson et al., 2017, 2018), which documented increased feeding efficiency under nCPAP compared to no NRS or to HFNC. We believe that the discrepancy between our studies is attributable to the specific study design required for videofluoroscopy assessment. Accordingly, during videofluoroscopy, lambs had to drink by themselves from our custom-made bottle-feeding device (Alain et al., 2021), conversely to our two initial studies in which the lambs were gently held by the same experimenter offering a bottle.

On days 13–14, sucking amplitude was greater during nCPAP than during HFNCcpap. This finding is somewhat in agreement with our recent study, which showed that HFNCcpap decreased feeding efficiency in full-term lambs (Alain et al., 2021).

First, we acknowledge that the use of thoracic compression to induce tachypnoea does not reproduce the pathophysiology of any common respiratory disease in the preterm. The relevance of studying tachypnoea, however, stems from the fact that it is virtually always present in preterm infants with bronchopulmonary dysplasia. In addition, it is the only standardisable way we found to impair respiration transiently, which was a must for our study. Second, our videofluoroscopy equipment did not allow us to use a rate higher than 12 frames•s^–1, which is slightly less than the recommended 15 frames•s^–1 in children (Hiorns and Ryan, 2006; Layly et al., 2020). This may have prevented us from recognising some laryngeal penetrations or tracheal aspirations. Third, the prone position with an extended neck—naturally adopted by newborn lambs while bottle-feeding—clearly does not reproduce the feeding position of a human newborn. Fourth, given that all lambs had been suckled from birth by their mothers, they had certainly developed some sucking–swallowing–breathing coordination prior to the experiment. Consequently, our results were obtained in a setting different from preterm infants, whose oral feeding is initiated under respiratory support. Fifth, although initially included in the experimental design, the effect of sex unfortunately could not be assessed due to the low number of female lambs born (4 females vs. 11 males). Lastly, although we unequivocally acknowledge that results obtained in preterm lambs cannot be taken as fully representing data in human preterms, many anatomical and functional characteristics of nutritive swallowing are common to both species. Anatomically, the upper aerodigestive tract is divided into an airway and a foodway with an overlapping epiglottis and soft palate in both species. The ovine larynx has an overall similar structure and is used for human surgical training (Isaacson et al., 2015, 2016; Ianacone et al., 2016). In both species, the cricopharyngeus muscle is the main component of the upper oesophageal sphincter (Samson et al., 2021). Functionally, there is similar immaturity of swallowing–breathing coordination during nutritive swallowing (frequent apnoeas during oral feeding) (Lau et al., 2003; Mizuno and Ueda, 2003; Gewolb and Vice, 2006). Similar vagovagal laryngeal chemoreflexes characterised by cardiorespiratory events can be elicited (Pickens et al., 1988; St-Hilaire et al., 2007).

Details of the chronic instrumentation have been described elsewhere (Samson et al., 2017; Alain et al., 2021). Briefly, on postnatal day 6, chronic surgical instrumentation was performed under local anaesthesia and included the insertion of (1) custom-built bipolar electrodes into both thyroarytenoid (EAta, a laryngeal constrictor) muscles for recording of swallowing activity; (2) a catheter into the left carotid artery to measure arterial blood gases (RapidLab 348, Siemens, Saint-Laurent, Canada); and (3) a transcutaneous catheter between the fifth and sixth tracheal rings to monitor tracheal-pressure variations.

Further instrumentation of the lambs was completed immediately before recordings and included (1) subcutaneous needle electrodes for electrocardiogram (ECG) recordings; (2) thoracic and abdominal elastic bands to monitor lung volume variations semi-quantitatively via respiratory inductance plethysmography (Ambulatory Monitoring, Ardsley, NY, United States); and (3) a pulse oximetry probe (LNOP YI reflectance sensor, Masimo, Irvine, CA, United States) at the base of the tail for continuous monitoring of SpO₂.

As previously described, nasal continuous positive airway pressure was delivered with the Infant Flow^® nCPAP system (Cardinal Health, Dublin, OH, United States), which delivered a variable-flow nCPAP of 6 cmH₂O through a custom-designed plastic nasal mask for newborn lambs (Supplementary Figure S1). A high nasal flow of 7 L•min^–1 of air was delivered with the Optiflow™ system (BC2755, Fisher & Paykel, Mississauga, ON, Canada) with small infant-size nasal cannulas. Lastly, given our past observation that bottle-feeding efficiency was enhanced under nCPAP (6 cmH₂O) but not under HFNC (7 L•min^–1) (Samson et al., 2018), we also studied a third type of NRS combining HFNC and a positive airway pressure of 6 cmH₂O (HFNCcpap). The latter was achieved with the Optiflow™ system at 7 L•min^–1 with small adult-size nasal cannulas and by reducing leaks at the nares with dental-impression material (EXAMIX^®, Henry Schein Canada, ON, Canada) to obtain an end-expiratory tracheal pressure of 6 cmH₂O (Alain et al., 2021). Lastly, for the control condition, the lambs were bottle-fed without any respiratory support and without a nasal mask. The air administered with all forms of NRS was heated and humidified. No oxygen was given to any of the lambs.

The lambs were fed without assistance from a custom-designed bottle-feeding device developed to allow videofluoroscopic studies without exposing the personnel to radiation (Supplementary Figure S2) (Alain et al., 2021). A saline-filled catheter connected to a pressure transducer (TSD104A pressure transducer, Biopac Systems Canada, Montreal, QC, Canada) was introduced in the teat of the bottle to record sucking activity (positive expression amplitude). The system allowed for adjusting teat height according to animal morphology.

All bottle-feeding sessions were performed under videofluoroscopy [Philips BV Pulsera C-Arm (Markham, ON, Canada)] to assess laryngeal penetration and/or tracheal aspiration. A mixture of 25% barium solution (Liquid Polibar Plus, E-Z-EM, Anjou, QC, Canada) and 75% milk was used to achieve a proper radiological contrast without modifying the texture, thickness, or colour of the milk (previously quantified in pilot experiments in 2 full-term lambs). The C-Arm was operated in sequential mode and set to 12 frames•s^–1 with a fixed X-ray energy setting in the 49–53 kVp range and a current intensity of 0.4–0.8 mA, as required for contrast in a given subject.

Physiological signals were transmitted wirelessly with our custom-designed radiotelemetry system (Samson et al., 2011) and continuously recorded (AcqKnowledge software, version 4.1, Biopac Systems Canada, Montreal, QC, Canada). The entire recording period was video recorded with a webcam, and an experimenter was present to note all events occurring during the recordings.

Following their birth, all preterm lambs were cared for with their mother in our animal quarters and were able to feed ad libitum on the ewe. Bottle-feeding was initiated on postnatal days 7–8 and repeated on days 13–14 to study the effect of maturation. The lambs were comfortably positioned prone in a sling with loose restraint. On the day of the recordings, one bottle of 40 mL of ewe milk—i.e., the maximum amount of milk that preterm lambs were able to drink during one bottle-feeding attempt, as determined in pilot experiments—was offered to each animal in the control condition (no NRS) and each of the 3 NRS conditions in random order (block randomisation) one day with and one day without respiratory impairment (randomly ordered days) under videofluoroscopic examination. A 3-h interval was systematically respected between two bottle-feedings (Alain et al., 2021). Videofluoroscopic recordings were continued for 20 s after bottle-feedings. The experimental conditions included no respiratory support (control), nCPAP at 6 cmH₂O, HFNC at 7 L•min^–1, and HFNCcpap. Respiratory impairment (tachypnoea) was induced by inflating a blood-pressure cuff around the thorax and the upper abdomen until the respiratory rate increased by 50%.

As extensively detailed previously (Samson et al., 2017, 2018; Alain et al., 2021), all physiological signals were analysed to quantify the safety and efficiency of bottle-feeding, as well as the SU–SW–BR coordination (rhythmic stability of feeding) of each bottle-feeding attempt in each of the three NRS and the control conditions, with and without tachypnoea. The data was analysed 20 s before bottle-feeding (=baseline), throughout all the bottle-feeding attempts and subsequently during 30 s.

The safety of bottle-feeding was quantified by computing the following variables: the number of heart-rate (HR) decelerations (defined by a percentage decrease in HR of 33% for a maximum of 5 s) and bradycardias (HR slowing >5 s), the minimal HR (min^–1), the total duration of cardiac inhibition (sum of all the HR decelerations and bradycardias for each reflex in seconds), the percentage decrease in HR (decrease in percentage from baseline HR to minimal HR during bottle-feeding), the minimal SpO₂ (%), the percentage decrease in SpO₂ (%), the number of coughs (defined by a distinctive audible sound combined with a brisk and short increase in tracheal pressure and EAta), the number of laryngeal penetrations (presence of milk in the laryngeal vestibule above the glottis), and tracheal aspirations (presence of milk below the glottis) (Ferrara et al., 2017). The videofluoroscopic recordings were analysed with a RadiAnt DICOM Viewer (version 5.0.0, Poznań, Poland)¹.

The rhythmic stability of feeding was quantified by computing the time interval between two sucks (SU–SU) and between two swallows (SW–SW), as well as the SU–SW and SW–breath (BR) intervals. The coefficients of variation [COV = standard deviation of the mean interval divided by the mean interval (Gewolb et al., 2001)] of SU–SU, SU–SW, SW–SW, and SW–BR intervals were also calculated. A lower COV value indicates a more stable rhythm (Da Costa et al., 2008). In addition, the percentage of feeding duration spent in apnoea (defined as at least two missed breaths with an amplitude 30% lower than baseline breathing (Samson et al., 2017) as well as the percentage of swallows occurring during an apnoea, was calculated.

The efficiency of bottle-feeding was quantified by computing the rate of milk transfer (mL•s^–1), the number of sucks (sucks•mL^–1) and swallows (swallows•mL^–1) needed to drink one mL of milk, and the mean positive pressure in the teat.

End-expiratory tracheal pressure was used to measure the positive pressure applied to the respiratory system by nCPAP, HFNC, and HFNCcpap (Supplementary Table S13). Lastly, the respiratory rate (RR), HR, mean SpO₂, and arterial blood gases (PaCO₂, PaO₂, and pH) were measured just before and 1 min after bottle-feeding.

All statistical analyses were performed with R V. 3.6.0 (The R Project for statistical computing, 2019, Vienna, Austria), with the help of the Department of Biostatistics of our research centre. A total of 37 dependent variables were evaluated. We first compared the impact of the three NRS conditions on the safety of bottle-feeding with and without tachypnoea. Similar comparisons were performed for efficiency and SU–SW–BR coordination. All analyses were initially conducted on the bottle-feeding periods without tachypnoea only and repeated for the bottle-feeding periods with tachypnoea. For each dependent variable, a generalised linear mixed-model analysis was performed. Tukey’s post hoc analysis was used for multiple comparisons of the mean differences between the conditions (a p-value < 0.05 was deemed significant). For each variable, the most representative distribution was selected based on the Akaike information criterion. Gamma, lognormal, or normal distributions were used for continuous variables, while Bayesian distribution was always the most appropriate one for discrete variables. The effect of maturation was tested using the same statistical model. Results are presented as median and first and third quartiles (Q1 and Q3).