Cohort characterisation
Seventy people with COPD (60 male, 68 ± 9y, BMI 25.5 ± 3.5, FEV1pp 48 ± 16, GOLD A-12, B-32, C-5, D-21) and fifty-eight sex and age matched healthy individuals (42 male, 67 ± 8y, BMI 27.6 ± 3.8, FEV1pp 103 ± 17) were included in this study. Detailed characteristics of participants are available in Table 1 and e-Table 1.
Table 1
Sociodemographic, anthropometric and clinical characteristics of participants included in the study. Comparisons between people with COPD and Healthy controls were conducted with unpaired t-test with Welch’s correction, Mann-Whitney U-test and Fisher’s exact test.
Characteristics
|
COPD
(n = 70)
|
HEALTHY (n = 58)
|
p-value
|
Age (years), mean ± SD
|
67.9 ± 8.7
|
67.0 ± 8.2
|
0.7
|
Male sex, n (%)
|
60 (86%)
|
42 (84%)
|
0.8
|
BMI (kg/m2), mean ± SD
|
25.5 ± 3.5
|
27.6 ± 3.8
|
0.001
|
Pack-Years, mean ± SD
|
42.2 ± 45.3
|
8 ± 21.0
|
< 0.0001
|
CCI, mean ± SD
|
3.7 ± 1.3
|
2.0 ± 1.0
|
< 0.0001
|
Medication for COPD, n (%)
|
70 (100%)
|
0 (0%)
|
|
Smoking Status, n (%)
|
Current Smoker
|
7 (10%)
|
2 (4%)
|
< 0.0001
|
Former Smoker
|
49 (70%)
|
10 (20%)
|
Never Smoker
|
14 (20%)
|
38 (76%)
|
GOLD Grade, n (%)
|
1
|
7 (10%)
|
n.a.
|
|
2
|
25 (36%)
|
n.a.
|
3
|
26 (37%)
|
n.a.
|
4
|
12 (17%)
|
n.a.
|
GOLD Group, n (%)
|
A
|
12 (17%)
|
n.a.
|
|
B
|
32 (46%)
|
n.a.
|
C
|
5 (7%)
|
n.a.
|
D
|
21 (30%)
|
n.a.
|
Long-term oxygen dependence, n (%)
|
11 (16%)
|
0 (0)
|
|
SpO2, mean ± SD (%)
|
94.4 ± 1.9
|
96.7 ± 1.7
|
< 0.0001
|
FEV1 (L)
|
1.3 ± 0.4
|
2.8 ± 0.6
|
< 0.0001
|
FEV1pp, mean ± SD
|
48.0 ± 16.4
|
103.0 ± 16.7
|
< 0.0001
|
FVC (L)
|
2.7 ± 0.6
|
3.4 ± 0.7
|
< 0.0001
|
Ratio FEV1FVC
|
48.7 ± 12.1
|
83.8 ± 8.7
|
< 0.0001
|
Number of exacerbations in the year before enrolment, n (%)
|
0–1
|
49 (70%)
|
n.a.
|
|
≥2 or 1 with hospital admission
|
21 (30%)
|
n.a.
|
Hospital admissions due to COPD, in the year before enrolment, n (%)
|
0
|
60 (86%)
|
n.a.
|
|
1
|
10 (14%)
|
n.a.
|
n (%): number of individuals in each group plus the corresponding percentage. mean ± SD: mean ± standard deviation. CCI: Charlson Comorbidity Index; BMI: Body Mass Index; GOLD Grade: 3– Severe; 4– Very Severe; GOLD Group: A– Less symptoms and low risk of exacerbations; B– More symptoms and low risk of exacerbations; C– Less symptoms and high risk of exacerbations; D– More symptoms and high risk of exacerbations; FEV1pp: forced expiratory volume in 1 second percentage of predicted; SpO2: peripheral capillary oxygen saturation. Comparisons between patients with COPD and Healthy controls were conducted with unpaired t-test with Welch’s correction, Mann-Whitney U-test and Fisher’s exact test.
|
Salivary microbiota composition and diversity is different between people with COPD and healthy controls
Principal coordinate analysis of pairwise distances (Weighted Unifrac) between healthy and people with COPD showed significant differences in microbiota composition between groups (PERMANOVA adjusted for PY, p=0.034) and captured 65% of total diversity (top three principal coordinates).
Microbiota of healthy individuals was composed of two major phyla, Firmicutes (40.6%) and Bacteroidetes (30.4%) (figure 1a). These were followed by Proteobacteria (16.3%), Fusobacteria (6.7%), Actinobacteria (2.5%) and six low abundant phyla (<3.5%). In terms of genera, Streptococcus (23%), Prevotella (24%) and Haemophilus (11%) were the most abundant. People with COPD showed a similar microbiota composition to healthy individuals, however differences in the relative frequencies of Bacteroidetes (26.5%) and Proteobacteria (22.3%) were observed as well as in genera Prevotella (18%) and Haemophilus (15%)
Differential abundant bacterial groups between people with COPD and healthy individuals were inferred with LEfSe and ANCOM. Both methods showed that healthy individuals were enriched in Treponema (Spirochaetes), Peptococcus (Firmicutes) and Peptostreptococcus (Firmicutes), whereas according to LEfSe, patients were enriched in genera from Proteobacteria and Firmicutes. Specifically, people with severe airflow obstruction showed an enrichment in Haemophilus, while those with moderate airflow obstruction were enriched in Granullicatella and Lachnoanaerobaculum (see supplementary Fig. 1 for the complete list of genera that differ between the groups).
Microbiota of people with COPD was significantly less diverse (Phylogenetic diversity - Alpha diversity, i.e., within individual diversity) than that of healthy individuals (figure 1d, Mann-Whitney U test, U=1275, p=0.0013). Similar differences were observed after adjusting for PY (ANOVA, F-value=10.89, p=0.002).
Salivary microbiota composition and diversity is poorly associated with clinical features
We next explored the relationship between the salivary microbiota and patients’ clinical features. Specifically, we queried whether different levels of airflow obstruction (GOLD grades) and severity of previous exacerbations and symptoms (GOLD groups) were associated with significant differences in microbiota diversity and composition.
Considering airflow obstruction, moderate patients (GOLD 1 & 2) bore a significantly distinct microbiota composition when compared with severe patients (GOLD 3 & 4) (PERMANOVA adjusted for PY, p = 0.002) but no significant differences were observed in alpha-diversity (Mann-Whitney U-test, U = 435, p = 0.12).
PCoA analysis separated A + B from C + D groups based on the severity of previous exacerbations but not A + C from B + D groups based on the severity of symptoms (PERMANOVA adjusted for PY (A + B vs C + D), p = 0.03, PERMANOVA adjusted for PY (A + C vs B + D), p = 0.06). Alpha-diversity was not significantly different among different levels of severity of previous exacerbations or symptoms (Mann-Whitney U test (A + B vs C + D), U = 420, p = 0.21; Mann-Whitney U test (A + C vs B + D), U = 392, p = 0.64).
No significant associations were found between alpha-diversity and pack-years, hospital admissions, long-term oxygen therapy, treatment with inhaled corticosteroids and SpO2 in people with COPD.
Salivary microbiota is associated with disease severity in people with COPD
In an effort to understand to what extent the oral microbiota is able to stratify COPD we performed a clustering analysis using the salivary microbial composition of patients. This analysis separated 90% of the individuals in two well supported clusters (“Cluster I” bootstrap node support (bns) = 74% and “Cluster II”, bsn = 84%; Fig. 2) which significantly differed in disease severity.
Cluster I aggregated all subjects with a history of recent severe exacerbation leading to hospital admission (Chi-square test, Z = 5.01, p = 0.025)) and 71% of the GOLD D (Chi-square test, Z = 1.98, p = 0.048). Two thirds of those under long term oxygen therapy or with heavier smoking history were also allocated to Cluster I. No other clinical parameters showed significant differences between the two clusters (supplementary table 2).
Microbiota composition was significantly different between the two clusters (PERMANOVA adjusted for PY, P = 0.001. Figure 2b.). Cluster I was enriched in patients dominated by Firmicutes or Proteobacteria, whereas cluster II was mainly represented by patients dominated by Bacteroidetes.
Microbiota diversity among patients (alpha diversity) was lower in Cluster I than in Cluster II (Fig. 2c. Mann-Whitney U-test, U = 271, p = 0.008). Similar differences were observed after adjusting for PY (ANOVA, F-value = 5.6, p = 0.006).
Regarding differentially abundant bacteria, both LEfSe and ANCOM distinguished Cluster I as particularly enriched in Streptococcus (Firmicutes) and detected Prevotella and Alloprevotella as responsible for the overabundance of Bacteroidetes in Cluster II (see supplementary Fig. 2 for the complete list of OTUs detected by LEfSe). Both methods further detected a significant enrichment of Dialister (Firmicutes) in Cluster II.
Logistic regression analyses were performed to quantify the risk afforded by the prevalence of Firmicutes, Proteobacteria and Bacteroidetes in the microbiota profile of people with COPD belonging to the two clusters. Furthermore, since three ASVs belonging to each of these phyla (Prevotella melaninogenica (Bacteroidetes), Haemophilus parainfluenzae (Proteobacteria) and Streptococcus sp. (Firmicutes)) were the main responsible for cluster segregation (Fig. 2b), the predictive power of their frequency was also inspected.
The combined frequency of Prevotella (Bacteroidetes) and Proteobacteria was found to be the best predictor of being GOLD D, (AUC = 87%), supplementary table 3 and Fig. 3a), i.e., patients with lower frequency of Prevotella and higher frequency of Proteobacteria were more likely to be severe.
Moreover, the odds ratio (OR) of 0.44 suggests a protective effect for increasing frequencies of Prevotella, while the OR of 2.83 suggests a risk effect for increasing frequencies of Proteobacteria.
Prevotella was the best predictor of recent severe exacerbation (leading to hospital admission) (AUC = 89%), which translated into a significantly higher risk for patients with low frequencies of this genus. The OR of 0.58, corroborated the protective effect (supplementary table 3 and Fig. 3b). Additionally, the frequency of Prevotella melaninogenica alone was also a good predictor for recent severe exacerbation (AUC = 86%) (supplementary table 3 and Fig. 3b), similarly the OR of 0.63 suggests a protective effect for higher frequencies of this ASV.
No significant associations were found considering Bacteroidetes, Haemophilus, Haemophilus parainfluenzae, Firmicutes, Streptococcus or Streptococcus sp. relative frequencies.