Patient Characteristics
Two hundred and fifty patients were enrolled in this study. The ascites of 7 patients were malignant ascites. Secondary peritonitis occurred in 6 patients, and 46 patients without complete data were excluded. One hundred and ninety one patients were enrolled, and their ascites samples were collected. Among them, 155 were male (81.2%), with an average age of 58.1 ± 9.1 years. Liver cirrhosis was caused by alcoholic hepatitis in 73 cases (38.2%), viral hepatitis B in 71 cases (37.2%), viral hepatitis C in 7 cases (3.7%), and other causes in 40 cases (21.0%). There were 41 suspected SBP cases (21.5%), 18 bacterascites cases (9.4%), and 132 no-AFI (69.1%). The demographic data are listed in Table 1.
Table 1
Clinical characteristics of enrolled patients.
| SBP (n=41) | Bacterascites (n=18) | No-AFI (n=132) | P ¶ |
Age (years, mean ± S.D.) | 58.6 ± 8.3 | 59.4 ± 10.8 | 57.8 ± 9.2 | 0.766 |
Gender (Male/Female) | 37/4 | 14/4 | 104/28 | 0.298 |
Etiology | | | | |
Alcohol | 17 (41.5) | 9 (50.0) | 47 (35.6) | 0.934 |
HBV | 15 (37.5) | 6 (33.3) | 50 (37.9) |
HCV | 1 (2.5) | 0(0) | 6 (4.5) |
HBV plus alcohol | 4 (9.8) | 1 (5.6) | 9 (6.8) |
Others | 4 (9.8) | 2 (11.1) | 20 (15.2) |
Complications | | | | |
Ascites 2/3 | 24/17 | 10/8 | 80/52 | <0.001 |
Gastrointestinal bleeding | 3 (7.5) | 2 (11.1) | 12 (9.0) | <0.001 |
Hepatic encephalopathy | 4 (10.0) | 2 (11.1) | 17 (12.9) | <0.001 |
HRS | 5 (12.5) | 1 (5.6) | 22 (16.7) | <0.001 |
Laboratory parameters | | | | |
ALT (IU/L), median (IQR) | 18.2 (8.1,28.3) | 17.4 (10.1,24.3) | 19.5 (11.7,31.6) | 0.297 |
AST(IU/L), median (IQR) | 36.5 (18.7,63.7) | 24.2 (15.8,49.8) | 39.6 (24.9,61.6) | 0.147 |
albumin (g/dL), mean ± SD | 30.2 ± 4.7 | 29.3 ± 4.4 | 29.7 ± 4.1 | 0.741 |
Total bilirubin (µmol/L), median (IQR) | 54.8 (15.4,168.5) | 34.4 (21.3,46.9) | 52.3 (26.6,99.8) | 0.282 |
PTA, mean ± SD | 65.7 ± 30.9 | 59.2 ± 16.8 | 58.7 ± 18.7 | 0.231 |
Serum creatinine (µmol/L), median (IQR) | 91.2 (63.8,179.5) | 84.4 (64.8,134.0) | 81.7 (59.5,120.3) | 0.422 |
Platelets (109/L), median (IQR) | 91.5 (63.4,141.3) | 53.2 (26.2,72.7) | 79.5 (45.5,114.8) | 0.008 |
Ascites WBC count (×106/L), median (IQR) | 1693 (765,3620)* | 220 (92,322) | 200 (130,620) | <0.01 |
Ascites PMN count (×106/L), median (IQR) | 1437 (1123,3587) | 40 (8,111) | 54 (28,94) | <0.01 |
Scores | | | | |
CTP | 10.1 ± 1.8 | 9.6 ± 1.4 | 10.0 ± 1.6 | 0.688 |
MELD | 12.9 ± 9.8 | 9.6 ± 5.2 | 11.9 ± 7.9 | 0.383 |
SD, standard deviation; IQR, interquartile range; HRS, hepatorenal syndrome; ALT, alanine aminotransferase; AST, aspartate aminotransferase; PTA, prothrombin activity; CTP, child-turcotte-pugh score; MELD, model for end-stage liver disease; PMN, polymorphonuclear neutrophils; WBC, white blood cell. AFI, ascitic fluid infection. |
P¶ value from Kruskal-Wallis test for continuous variables or Fisher’s exact test for discrete variables comparing patients with SBP to patients with bacterascites and No-AFI. |
*P<0.05 |
Analysis And Valuation Of Ddpcr Method
Results showed that the clustering effect of ddPCR was best when the primer probe concentration was 400/200 nmol/L and the annealing temperature was 60 ℃(Supplementary Table S1, Fig. 1). Representative Gram-positive and Gram-negative bacteria were selected for a specific probe test, which indicated that the probe could clearly distinguish Gram-positive or Gram-negative bacteria (Fig. 2a, b). To determine whether the probe had cross interference in distinguishing Gram-positive and Gram-negative bacteria, we mixed E. feacium and E. coli in different concentrations (100:1, 1:100, 1:1) and tested them. The results showed that the corresponding quantitative level of bactDNA were 4.3: 2.1 log copies/µl, 2.0 : 4.0 log copies/µl, and 2.2 : 2.3 log copies/µl, indicating that reaction systems with different concentrations could be accurately classified and quantified (Fig. 2c, d, e).
To determine the ddPCR detection limit, linearity, and repeatability, 9 kinds of bacteria were spiked in a mixed system and subsequently detected by ddPCR. Serial dilutions of the above bacteria at known concentrations showed a good linearity (R2 = 0.97- 0.99; Supplementary Fig. 2 ), with three replicates at each dilution (1 - 5 log copies); the ddPCR detection limit was approximately 20 - 45 copies/µl for bacterial strains. Compared with traditional qPCR methods, the ddPCR showed better linearity and lower detection limits (Supplementary Fig. 3).
Clinical Evaluation Of An Improved Quantitation Of Ddpcr Method
Benzonase endonuclease digestion was performed on the samples before detection, so that all detected DNA came from live bacteria. We selected randomly 54 cirrhotic ascites samples (13 SBP, 15 bacterascites, 26 no-AFI samples). Results showed that compared to the extraction method of None-dependent assay, Benzonase-dependent assay had a significant difference (P < 0.001) and even lower quantitation of bactDNA (1.75 vs 1.5 copies/µl, Fig. 3a, b). And then we found that this difference between the two methods of extraction was mainly the extraction of gram-negative bacteria from bacterascites and no-AFI (Fig. 3c), which were conducive to distinguish the interference of cell-free DNA fragments and the real bacterial infection with low loads. After the depletion of DNA fragments, the area under the ROC curve were as follows: bactDNA 0.98 (95%CI, 0.94-1.00), Gram-positive bactDNA 0.91 (95%CI, 0.84-0.99), Gram-negative bactDNA 0.95 (95%CI, 0.88-1.00; Fig. 3d), which indicated that DNA treatment with benzonase may have a more diagnostic value.
The clinical conditions of cirrhotic patients with ascites are complex and varied, and the change of ascites volume may also affect the judgment of the absolute numbers of bacteria. We used the ultrasonic three-point method to evaluate and eliminate the influence of ascites volume and found that there was no significant variation in bactDNA quantitation between the copies per 1 µl ascites and the total copies of bactDNA combined with ascites volume (P > 0.53, Fig. 3e). Compared with bactDNA quantification in different ascites samples, the quantitation of bactDNA in the ascites of SBP patients (total 2.8, gram-positive 1.9, gram-negetive 2.5 log copies/µl) were significantly higher than that of bacterascites (1.7, 1.3, 1.5 log copies/µl) and no-AFI (2.0, 1.4, 1.8 log copies/µl, P<0.001, Fig. 3e, Supplementary Table S2), while the bactDNA of bacterascites and no-infectious ascites had no significant differences.
The enhanced sensitivity and specificity of bactDNA facilitated SBP diagnosis
Subsequently, we plotted ROC curves at varying bactDNA quantification corresponding to SBP analysis. Results from 191 samples showed that the cutoff value of bactDNA quantification was 103.2 copies/µl comparing to the diagnosis of SBP. For patients infected with Gram-positive bacteria and Gram-negative bacteria, the cutoff values were 37.3 and 68.6 copies/µl, respectively (Fig. 4a).
At the optimal Yueden’s index derived from the ROC curve, the sensitivity and specificity of total bactDNA quantitation comparing to the composite clinical standard were 80.5% (95%CI, 67.8%-93.2%) and 95.3% (95%CI, 91.1%-98.3%), the positive percentage agreement (PPA) and negative percentage agreement (NPA) were 82.5% and 94.7%, respectively. And the diagnostic test revealed sensitivity 72% (95%CI, 53.1%-90.9%), specificity 93.9% (95%CI, 90.3%-97.6%), PPA 64.3% and NPA 95.7% of SBP caused by Gram-positive bacteria, and sensitivity 91.3% (95%CI, 78.8%-100%), specificity 89.3% (95%CI, 84.6%-94.0%), PPA 53.8% and NPA 98.6% of SBP caused by Gram-negative bacteria (Fig. 4b). Additionally, the agreement of bactDNA positive results with culture, PMN, composite diagnostic standard positive results was 56.8% (95%CI, 40.0%-73.5%),71.4% (95%CI, 53.6%-89.3%), and 80.5% (95%CI, 67.8%-93.2%). The negative results agreement among them was 88.3% (95%CI, 83.2%-93.4%), 87.7%(95%CI, 82.6%-92.8%), and 95.3% (95%CI, 91.1%-98.3%),respectively (Table 2). Therefore, as a novel tool, bactDNA based on ddPCR greatly improves microbial diagnosis in SBP.
Table 2
Positive and Negative Agreement of quantitation of bactDNA of SBP versus culture, PMN, and composite diagnostic standard
Outcome (n=191) | Total bactDNA-positive | Total bactDNA-negative | Agreement(%) |
Traditional culture-positive (n=37) | 21 | 16 | 56.8 |
Traditional culture-negative (n=154) | 18 | 136 | 88.3 |
PMN ≥250/mm3 (n=28) | 20 | 8 | 71.4 |
PMN <250/mm3 (n=163) | 20 | 143 | 87.7 |
Composite diagnostic standard-SBP (n=41) | 33 | 8 | 80.5 |
Composite diagnostic standard-no-SBP (150) | 7 | 143 | 95.3 |
*PMN polymorphonuclear; SBP spontaneous bacterial peritonitis. |
BactDNA quantitation in patients with symptoms and PMN count <250/mm 3 can sensitively distinguish patients with suspicious infection
The data showed that 163 out of 191 samples had PMN less than 250 cells/mm3, of which 13 patients were consulted by 2 infectious physicians and 1 hepatologist to consider the diagnosis of SBP and were treated with empirical antibiotics (see Supplementary Table S3). After treatment, the patients’ symptoms improved and was clinically confirmed as SBP. In 13 SBP patients with a PMN < 250 cells/mm3, the quantitation of bactDNA in ascites were significantly higher than those in no-SBP patients (2.7, 2.2, 2.1 log copies/µl vs 1.7, 1.1, 1.5 log copies/µl, P < 0.001, Fig. 4c, Supplementary Table S4). Notably, we found that 7 patients with bacterascites were all culture-positive (Fig. 4d), including Enterococcus faecalis, Enterococcus faecium, Staphylococcus haemolyticus, Corynebacterium, which were consistent with previous study in bacterascites[6]. According to the counts of PMN from both culture and ddPCR, those of gram-positive bacteria were lower than gram-negative bacteria (Fig. 4e). On the view of our data, the cutoff value of PMN count was 192 cells/mm3 with 73.2% (95%CI, 59.4%-87.2%) sensitivity and 98.6% (95%CI, 97.3%-100%) specificity compared to composite diagnostic standard. Therefore, we hypothesized that a PMN threshold of 250/mm3 was too high for the diagnosis of SBP, especially for Gram-positive infections.