Retrospective imaging cohort
The clinical utility of computed tomography pulmonary artery angiography (CTPA) has been verified for monitoring of COVID-19 in our clinical setting, consistent with recent reports by other groups.17 We retrospectively analyzed all consecutive CTPA examinations acquired at Karolinska University Hospital in Huddinge, Stockholm, Sweden, in patients with reverse transcription-polymerase chain reaction (RT-PCR) confirmed SARS-CoV-2 infection. In total, 339 CTPA examinations were performed in 289 adult patients. Motion artifacts or other technical issues prohibited pulmonary artery (PA) diameter measurements in 47 examinations, as measured by the gold standard reader, TG, resulting in 292 analyzed scans. The mean PA diameter was 29.2 ± 4.3 mm (mean ± SD), with 174 of 292 (60%) demonstrating abnormally wide PA diameter of ≥28 mm, suggestive of pulmonary hypertension.18 Pulmonary macro-thrombosis/embolism was present in 56 of 336 examinations (17%, 3 scans were not interpretable due to technical quality). We then analyzed a subset of the CTPA cohort that had also undergone echocardiography (n = 50). Echocardiography indicated high PA pressure: maximal tricuspid regurgitation velocity 3.0 ± 0.43 m/s (N=38; normal reference ≤ 2.8); estimated systolic pulmonary artery pressure 47 ± 14 mmHg (N=38; normal reference < 36); right ventricular outflow tract acceleration time 57 ± 43 ms (N=17; normal reference > 100).19,20 Maximal tricuspid regurgitation velocity and estimated systolic pulmonary artery pressure were not measurable in 12 patients due to insufficient tricuspid valve regurgitation Doppler signal.
These findings were corroborated by invasive pressure measurement obtained in a SARS-CoV-2 RT-PCR-positive 58-year-old male ICU patient. A PA-catheter was placed on ICU day 25 and recorded systolic and diastolic PA pressures of 51 ± 2.9 / 11 ± 3.2 mmHg (mean ± SD) and SvO2 of 62.4 ± 1.3 % (mean ± SD), both during the first hour after placing the catheter. Upon admission to the ICU, the patient’s clinical chemistry values were PaO2 of 8.0 kPa (reference range 8.0-13 kPa) and PaCO2 of 5.1 kPa (reference range 4.6–6.0), IL-6 of 95 ng/L (reference range <7), TNF-α of 9.8 ng/L (reference range <12), C-reactive protein of 66 mg/L (reference range <3), AST of 0.53 μkat/L (reference range <0.76 D-dimer of 0.26 mg/L FEU (age-adjusted cut-off <0.58 mg/L FEU).
Clinical chemistry characteristics of imaging cohort
We retrospectively collected clinical chemistry data in the CTPA cohort based on the blood sampling closest to the exam date, within ± 3 days from the first CTPA examination. The median of D-dimer was 1.13 (lower quartile 0.66, upper quartile 3.0, reference range <0.5 mg/L FEU, n=257), fibrinogen 6.3 (lower quartile 4.6, upper quartile 7.4, reference range 2–4.2 g/L, n=91), interleukin-6 76 (lower quartile 38, upper quartile 151, reference range <7 ng/L, n=147) and TNF-α 13 (lower quartile 10, upper quartile 19, reference range <12 ng/L, n=78).
MRI lung perfusion
A clinical MRI lung perfusion scan had been performed in a SARS-CoV-2 RT-PCR-positive 61-year-old male on ICU day 22 with an initial presentation to the ICU that included a PaO2 of 7.6 kPa (reference range 8.0–13 kPa) and a PaCO2 of 3.6 kPa (reference range 4.6–6.0 kPa) by arterial blood gas; serum measurements of IL-6 154 ng/l (reference range <7 ng/L), TNF-α 10.2 ng/L (reference range <12 ng/L), C-reactive protein 242 mg/L (reference range <3 mg/L), AST 1.08 μkat/L (reference range <0.76 μkat/L) and D-dimer 0.94 mg/L FEU (age-adjusted cut-off <0.61 mg/L FEU). In this MRI lung perfusion study, the maximum contrast bolus concentration in the PA came at 15 seconds after injection and the maximum contrast bolus concentration reached the aorta after 20 seconds (Fig. 1). Using these values, we calculated a time-to-peak (TTP) map of the lungs, describing how long the injected contrast took to arrive at a specific area. The in vivo perfusion map presented in Fig. 1, demonstrates late contrast arrival, oftentimes later than the aortic arrival time. To further understand these paradoxical values, we measured regions of interest in the normal-appearing lungs without infiltrates, as identified by T2-weighted anatomical images.
The peripheral regions of the lungs did not receive a contrast bolus even 10 seconds after the contrast had peaked in the aorta. We manually segmented the lung and calculated the fraction of the lung with contrast peak, or no contrast bolus at all, and found a ratio of 44% dysfunctional lung as defined by peak contrast enhancement after the aorta, suggestive of microvascular occlusion.
Summary of clinical results
We found multimodal evidence of raised pulmonary artery pressure confirmed by invasive pressure monitoring and a lack of blood perfusion in functional MRI. In a cohort of patients undergoing CTPA examination, we found no clear evidence of cytokine levels that are indicative of cytokine release syndrome, despite indication bias of sampling that would potentially skew the cohort characteristics towards those patients most ill. To further explore the implications of these findings, we turned to large animal experiments.
Large animal infusion of supraphysiological levels of angiotensin II
It seemed highly probable that at least one part of the COVID-19 pathology involved vasculature disturbances and the RAAS and we, therefore, designed a hypothesis-generating experiment where we infused ANGII in three sedated swine targeting a systolic arterial pressure of 150 mmHg. Within five minutes, arterial and PA blood pressures started to climb (Fig. S3). All three swine reached PA-systolic pressures exceeding 30 mmHg during the experiment and one swine died of acute right ventricular heart failure with extensive pulmonary thrombosis/embolism. ANGII infusion further induced a rapid decline in SvO2, falling below 50% in all swine. Trends in arterial PaO2 were downward and arterial PaCO2 was trending upwards (Fig S4). While performing an autopsy
on Swine #1, we found a 13 cm long pulmonary thrombus. To exclude embolization, we started screening for deep venous thrombosis by ultrasonography but none were detected (Table S1). Samples for histology were acquired from macroscopically wedge-shaped blood discolored areas. After hematoxylin and eosin staining, we observed thrombotic material in small vessels, thickening of alveolar septa and debris in the alveolar sacs.
We performed bleeding time assessments in Swine #2 and #3 with baseline bleeding times of 285 seconds and 255 seconds respectively. These were reduced to < 120 seconds after 90 minutes and remained low for the remainder of the experiment. The von Willebrand factor activity (GP1bA) increased directly after the initiation of the infusion. In this hypothesis generating experiment, D-dimer increased and paradoxically, fibrinogen also increased or remained unchanged despite increased D-dimer. IL-6 and TNF-α remained below the level of detection in all swine during the experiment. Osmolality successively increased in all three swine by > 11 mosmol/kg. All clinical chemistry data is individually represented in table S1.
Infusion of supraphysiological levels of ANGII, a model akin to “accelerating the RAAS” seems to produce a remarkably unhealthy state in the swine. We hypothesize that COVID-19 patients have lost the “break” in the RAAS. Therefore, we performed additional experiments blocking ACE2 by MLN-4760.
Large animal blocking of ACE2 with low rate infusion of angiotensin II
We injected the ACE2 blocker MLN-4760 and started an infusion of ANGII in three additional swine. Increases in systemic and pulmonary artery pressures were not as high as in the supraphysiological infusion group (Fig. S3). Surface body temperature was the same, trends in PaO2 were downward and PaCO2 was trending upwards (Fig. S4). Macroscopically, large wedge-shaped lung areas were found in the group receiving supraphysiological ANGII, highly suggestive of a more malignant model of disease by blocking ACE2. To further explore our hypothesis, we then combined the MLN-4760 blocker with the supraphysiological ANGII infusion regime. This leads to a severely malignant model with severe elevation of both systemic arterial and especially pulmonary arterial pressures, that would not be long term compatible with life (Fig. S3).
Treatment Screening in a Large Animal Model
To address the most common patient and lend mechanistic support to some of the on-going clinical trials we decided that the MLN-4760 blocker coupled with low dose infusion of ANGII might represent a clinical situation that was not as malignant as in the COVID-19 ICU patients, but could still simulate a COVID-19 patient in need of hospital intervention. Hereafter, MLN-4760 coupled with low dose infusion of ANGII will be referred to as “the animal model”. We decided to design a small study to test treatment with oral administration of 200 mg losartan, through an orogastric tube, combined with subcutaneous administration of 10,000 IU of low molecular weight heparin.
We performed MRI lung perfusion and used the same definition for peripheral perfusion deficit as in the patient, i.e. a contrast bolus peak later than the aorta or no peak at all (Fig. 1). The untreated model exhibited a significantly greater fraction of non perfused lung at 17%, compared to the treated group at 9% (n=6, SD untreated 1.5%, SD untreated 1.7%,
0.95 CI for difference 5–12, p=0.003, two-tailed t-test, assumed equal distribution). While the placebo group suffered a significant decrease in O2-saturation in the arterial blood gases over the course of the experiment (-0.54/hour), the treated group experienced no decline
(Fig. S4). The estimated difference between the groups was statistically significant (n=6, difference between groups: 0.53/hour, .95 CI: 0.096 – 0.96, p=0.017, Linear mixed error-component model, model coefficients in table S2). There were no significant differences in PaO2.
Furthermore, the PA diameter, as measured by MRI, was larger in the untreated model; 2.41 mm (SD ± 0.24) compared to 2.21 mm (SD ± 0.24) in the treated group. This difference was not significant (n=6, 0.95 CI for difference -0.3 – 0.72, p=0.32, two-tailed t-test, assumed equal distribution). These findings seem particularly interesting when considering the clinical CTPA cohort and echocardiography sub-cohorts with COVID-19, where evidence of increased pulmonary pressure, including widened PA-diameter, was observed.
During the initial hours of the animal model, no clear physiological difference could be observed between treated and untreated groups, but after approximately 2 hours systemic pressures started to return to baseline and pulmonary pressures never reached as high (Fig. S3). The estimated systolic PA pressure decreased significantly faster for the treated group compared to the untreated group which showed a trend towards higher systolic PA pressures during the intervention (n=6, -0.44/h, 0.95 CI: -0.84 – -0.040, p=0.03, linear mixed error-component model, model coefficients in table S2).
D-dimer increased in both groups, by on average 0.027 mg/L FEU per hour in placebo and 0.022 mg/L FEU per hour in the treated group. Notably, and congruent with the results from our hypothesis generating experiment, fibrinogen also increased for both groups, by 1.07 mg/L per hour in placebo and 0.47 mg/L per hour in the treated group (Table S1). Bleeding times were also reduced as in the supraphysiological ANGII infusion groups.
Echocardiography of two swine, one treated and one without treatment showed significant abnormalities in the latter, suggestive of pulmonary hypertension, with the same functional pattern as was observed in patients. The swine without treatment showed right ventricular (RV) enlargement, interventricular septal flattening, tricuspid regurgitation peak gradient > 30 mmHg, RV free wall hypokinesia and evidence of RV dysfunction.
We performed extensive histological analysis in both groups, and could confirm the same type of pulmonary damage as in the supraphysiological ANGII infused swine, predominated by a thickening of alveolar walls and debris deposition (Fig. 2). We could further discern marked end-organ damage with micro-necrotic changes in the small intestine and liver, as well as the kidney (Fig. 2).
Summary of preclinical results
We demonstrate that both “acceleration” and “loss of brake” of the RAAS, will lead to a pathophysiological phenotype closely resembling that found in COVID-19. Furthermore, we use this model to test pharmacological treatments currently under evaluation in clinical trials.
We show significant improvement in lung perfusion, as measured by MRI, O2-saturation in the arterial blood gases and PA-pressures providing mechanistic support for therapeutic strategies involving the RAAS and coagulation.