A Longitudinal Trans-Kingdom Gut Microbial Approach towards Decompensation in Outpatients with Cirrhosis

Abstract

Objective:

First decompensation development is a critical milestone that needs to be predicted. Trans-kingdom gut microbial interactions, including archaeal methanogens may be important targets and predictors but a longitudinal approach is needed.

Design:

Cirrhosis outpatients who provided stool twice were included. Group 1: Compensated, group 2: 1-decompensation (decomp), group 3: >1-decompensationwere followed and divided into those who remained stable or decompensated. Bacteria, viral & Archaeal presence, α/β diversity & taxa changes over time adjusted for clinical variables were analyzed. Correlation networks between kingdoms were analyzed.

Results:

157 outpatients (72 Group-1, 33 Group-2, 52 Group-3) were followed and 28–47% developed outcomes. Baseline between those who remained stable/developed outcome: While no α/β diversity differences were seen, commensals were lower and pathobionts were higher in those who decompensated. After decompensation: those experiencing their first decompensation showed greater decrease in α/β-diversity, bacterial change (↑Lactobacillus spp, Streptococcus parasanguinis and ↓ beneficial Lachnospiraceae & Eubacterium hallii ) and viral change (↑Siphoviridae, ↓ Myoviridae) versus those with further decompensation. Archaea: 19% had Methanobacter brevii., which was similar between/within groups. Correlation networks: Baseline archaeal-viral-bacterial networks were denser and more homogeneous in those who decompensated versus the rest. Archaea-bacterial correlations collapsed post-first decompensation. Lactobacillus phage Lc Nu and C2-like viruses were negatively linked with beneficial bacteria.

Conclusion:

In this longitudinal study of cirrhosis outpatients the greatest trans-kingdom gut microbial changes were seen in those reaching the first decompensation, compared to subsequent decompensating events. A trans-kingdom approach may refine prediction and provide therapeutic targets to prevent cirrhosis progression.

BACKGROUND:

With the changing demographics of patients with cirrhosis (HCV eradication, greater NAFLD and alcohol and aging population), the dynamics of predicting first and further decompensation are changing[1, 2]. This has several consequences for planning patient selection for trials designed to prevent the all-important first decompensation as well as enhancing outcome prediction[3, 4].

The gut microbiome is a major contributor towards hepatic encephalopathy (HE) and spontaneous bacterial peritonitis (SBP)[5, 6, 7]. However, the impact of the microbiome on development of all-cause decompensation as a whole is unclear[8, 9]. The current literature in cirrhosis and microbiota are mainly comprised of short-term trials of microbially-modified therapies and projections of data on follow-up without longitudinal collections[10, 11]. In addition to bacteria, cross-sectional studies of virome in cirrhosis have shown distinct alterations, but this again has not been studied longitudinally[12, 13]. Furthermore, the role of archaea, whose methanogenic properties could impact HE through slowing down intestinal transit, have not been studied in patients with cirrhosis[14, 15]. Therefore, the interaction of viruses and archaea with bacteria need to be further evaluated in a longitudinal, long-term context.

We hypothesized that bacteria, archaea, and viruses in the gut microbiota of stable outpatients are distinct between those who develop one decompensation event compared to those who develop further decompensation, or more than one event of decompensation compared to patients who have a stable clinical course.

METHODS

Subjects: We enrolled patients with cirrhosis prospectively after informed consent and IRB approval. Patients were followed at least every 6 months for decompensation. Stool samples were collected in the groups that developed initial/further decompensating events. These were balanced with patients whose clinical course had remained stable during that same interval, who were brought in for stool collection. We only included patients >18 years, who could consent, provide samples, and did not have recent hospitalization or alcohol-associated hepatitis (within 90 days). Cirrhosis was defined by liver biopsy, evidence of decompensation, presence of varices on endoscopy or radiology in a patient with chronic liver disease. Patients were divided into three groups: Group 1: compensated, Group 2: 1-decompensation, and Group 3: >1-decompensation. Decompensation was defined as ascites, HE, or prior variceal bleeding[16]. Repeat stool samples were only collected at least a month after the latest decompensating event.

Data and stool collection: we collected demographics, cirrhosis details (MELD score, ascites, variceal bleeding, SBP), medications and laboratory values. Changes in these events and medications between the baseline and the second event were also recorded.

Microbial extraction and analysis (supplement):

Bioinformatics:

Broadly, we studied changes in viral, bacterial, and archaeal α-diversity (diversity within a sampled environment), β-diversity (diversity between communities), and differential microbial features at baseline and longitudinally between/within groups who remained stable or developed further outcomes. These were adjusted for clinical parameters (age, gender, MELD score, medications, and alcohol-associated etiology). Finally, correlation networks, which define relationships between groups in complex systems, were analyzed. α-diversity: After baseline comparisons, we defined changes in α-diversity between those who developed outcomes/not and using the Wilcoxon-rank sum test. We also assessed α-diversity between those who developed their first decompensation and those who remained stable after 1 decompensation. β-diversity After baseline comparisons, we tested Wilcoxon rank sum tests within subjects and then between groups using Pre and Post timepoints with PERMANOVA-FL (Permutational Multivariate Analysis of Variance)[17]. PERMANOVA compares samples between two or more groups using β-diversity metrics to determine if the microbial communities are significantly different between groups of interest.

Differential microbial features: At baseline we first determined if we could differentiate the microbiota composition between those who remained stable versus those who developed decompensating events in the future in each group using Biomminer DESeq2 analysis [18, 19]. Further, Wilcoxon rank sum testing was used to determine if there were changes within individual subjects’ microbiome using centered Log-Ratio transformation (CLR) abundance of microbiota. This involves dividing the count abundance of each feature in a sample by the geometric mean of count abundances in that sample, followed by taking log of that ratio.DESeq2 was also performed between the three groups of patients post-decompensation who changed over time to see if first, second, or subsequent decompensations would have specific microbial signatures.

Archaea Methods:

Presence of Archaea (≥10 reads mapped) was compared using Fisher’s Exact test at baseline and at all timepoints. Also, we used a hurdle model (supplement) to determine changes in Archaea in stable versus unstable groups.

Correlation network analysis:

Was performed using R with published methods[20] to determine linkages between viruses, bacteria and archaea at baseline and study-end between and within groups. Only those with r>0.5/<−0.5 and p<0.05 were included. Correlation network characteristics, which include number of features (bacteria, viruses, or archaea) that are linked to each other, density, clustering, and heterogeneity of the linkages, and centering of the networks are all measures of how closely features or nodes in the correlation networks are interconnected[20]. These measures are used to compare correlation networks to determine differences in linkages between trans-kingdom groups in those are stable versus unstable and before/after decompensating events.

Patients and public:

were not involved.

RESULTS:

Patient characteristics and changes:

We included 157 outpatients with cirrhosis who gave stool twice between September 2015 to March 2021 (Table 1), with 72 group 1, 33 group 2 and 52 group 3. The median (IQR) between first and second sampling for those who developed first decompensation was 435 (429) days, which was matched with those who remained stably compensated and had stools collected 426 (321) days apart. For those with 1-decompensation at baseline, the second sample was collected 452 (409) days post-enrollment in those who developed their second decompensation, which was matched with those who had their samples collected 421 (391) days apart. For those with >1-decompensation at baseline, further decompensation occurred 148 (268) days which were compared to those who remained stable over 209 (206) days apart. During this time period 29% (n=21) of Group 1 decompensated, 46% of Group 2 (n=15) developed a further decompensation and 37% (n=19) of Group 3 developed further decompensation (Figure 1). At baseline, we observed that >1-decompensated patients who developed further decompensation were younger vs. those who remained stable; no other significant differences were found.

Table 1:

Baseline characteristics of the three groups

	Group 1 (never decompensated) (n=72)		Group 2 (one decompensating event at baseline) (n=33)		Group 3 (>1 decompensating event at baseline) (n=52)
	Stable (n=51)	Unstable (n=21)	Stable (n=18)	Unstable (n=15)	Stable (n=32)	Unstable (n=19)
Age	59.4±7.5	59.2±5.9	59.6±7.4	58.7±4.9	63.7±5.7	58.6±9.1*
Gender	36	18	11	10	27	17
Diabetes	20	9	7	5	17	7
Laboratory values
MELD score	8.9±3.3	9.20±2.6	10.78±2.7	12.7±2.9	13.0±4.4	12.7±2.3
Child score	5.16±0.37	5.85±1.01	6.97±0.92	7.63±1.79	8.00±1.69	8.21±1.11
Serum albumin	3.94±0.40	3.46±0.65	3.49±0.39	3.25±0.33	3.26±0.52	3.05±0.57
Serum sodium	140.12±2.40	138.57±2.62	137.50±4.57	135.73±4.46	136.75±7.06	134.42±5.75
Serum creatinine	0.94±0.27	0.94±0.28	0.923±0.28	0.80±0.25	1.07±0.33	1.16±0.41
INR	1.11±0.12	1.12±0.13	1.26±0.13	1.35±0.20	1.44±0.41	1.34±0.27
Serum bilirubin	0.82±0.48	1.11±0.64	1.58±1.29	2.12±1.46	2.15±3.52	1.71±0.84
WBC count	5.89±1.76	5.03±1.25	5.22±2.36	4.49±2.08	5.17±2.18	5.37±2.09
Hemoglobin	14.24±1.48	13.71±1.98	13.10±2.02	11.65±1.68	11.61±2.00	11.56±2.24
Platelet count	164.3±70.7	117.6±52.5	125.6±60.1	115.2±41.8	112.1±60.1	134.9±122.6
Complications
Ascites	0	0	10	11	26	18
HE	0	0	13	6	30	15
Variceal bleeding	0	0	0	2	15	9
Prior SBP	0	0	0	0	1	1
Medications
Lactulose	0	0	13	6	28	15
Rifaximin	0	0	9	3	27	12
PPI	19	10	9	10	22	13
Beta-blockers	20	11	7	4	18	6
Statins	18	6	4	4	9	1
SBP prophylaxis	0	0	0	0	1	1

No significant differences within groups between those who developed future complications versus those who remained stable apart from lower age in unstable subjects with >1 decompensation at baseline.

^*p=0.05

Figure 1:

Flow of subjects during the study using a flow chart (1A) and Sankey plot (1B)

Clinical changes on follow-up:

Most decompensations were due to the development of HE or ascites (Table 2). In Group 1, 9 patients developed ascites, 4 variceal bleeding and 10 HE (3 had concurrent HE and variceal bleeding). From Group 2, 2 patients developed new ascites, 4 new HE, 1 new SBP, and 7 had additional HE episodes. In Group 3 there were 2 patients with new variceal bleeding, 3 new HE episodes, 3 new SBP and 4 with additional HE episodes. Child score increased in those with first decompensation versus baseline; rest of the labs were similar. Rifaximin was initiated in 9 of the 21 new decompensated patients 99±47 days before second sampling. Lactulose was initiated in 12 of 21 new decompensated patients and was initiated 86±31 days before second sampling. It was continued in the remaining Group 2 and 3 patients. SBP prophylaxis was maintained in two patients in Group 2A, initiated in one patient post-SBP after their second decompensation (59 days pre-second sampling), maintained in one person in those were stable >1-decompensation. In >1-decompensation end it was initiated in 3 patients post-SBP (78±24 days pre-second sampling) and maintained in one person. Five patients were started on PPI after their first decompensation (96±63 days pre-second sampling); in the remaining groups the PPI use was maintained. Statin and beta-blocker use were similar (Table 2).

Table 2:

Changes in clinical criteria over time in stable and unstable patients.

	Group 1 (compensated) (n=72)				Group 2 (one decompensating event at baseline) (n=33)				Group 3 (>1 decompensating event at baseline) (n=52)
	Stable (n=51)		Unstable (n=21)		Stable (n=18)		Unstable (n=15)		Stable (n=32)		Unstable (n=19)
	Base	End	Base	End	Base	End	Base	End	Base	End	Base	End
Labs
MELD score	8.7±3.3	8.8±3.2	9.2±2.6	10.2±4.0	10.8±2.7	10.8±2.8	12.7±2.9	12.4±3.1	13.0±4.4	12.1±3.4	12.7±2.3	13.9±2.8
Child score	5.16±0.37	5.22±0.50	5.85±1.01	7.05±1.72*	6.44±0.92	6.56±1.04	7.93±1.79	7.80±1.47	8.00±1.69	7.97±1.58	8.21±1.11	8.37±1.40
Serum albumin	3.94±0.40	3.96±0.36	3.46±0.65	3.43±0.74	3.49±0.39	3.39±0.45	3.25±0.33	3.20±0.45	3.26±0.52	3.26±0.53	3.05±0.57	2.92±0.50
Serum sodium	140.1±2.4	140.1±2.5	138.6±2.6	139.8±2.8	137.5±4.6	139.2±4.2	135.7±4.5	134.2±7.9	137.8±4.1	138.8±2.8	134.4±5.8	135.9±5.2
Serum creatinine	0.94±0.27	0.99±0.28	0.94±0.28	1.09±0.3*	0.92±0.28	0.90±0.28	0.80±0.25	0.87±0.27	1.07±0.33	1.09±0.39	1.16±0.41	1.15±0.44
INR	1.11±0.12	1.12±0.12	1.12±0.13	1.20±0.21	1.26±0.13	1.30±0.17	1.35±0.20	1.37±0.19	1.44±0.41	1.35±0.22	1.34±0.27	1.39±0.19
Serum bilirubin	0.82±0.48	0.81±0.57	1.11±0.64	1.28±0.96	1.58±1.29	1.51±1.33	2.12±1.46	2.32±1.34	2.15±3.52	1.68±1.09	1.71±0.84	1.88±0.98
WBC count	5.89±1.76	5.98±2.00	5.03±1.25	5.41±2.18	5.22±2.36	5.32±2.35	4.49±2.08	4.75±2.41	5.17±2.18	5.42±1.91	5.37±2.09	5.67±2.09
Hemoglobin	14.24±1.48	14.13±1.77	13.71±1.98	12.95±2.24	13.10±2.02	13.04±1.84	11.65±1.68	11.97±1.82	11.61±2.00	12.10±1.99	11.56±2.24	11.05±1.77
Platelet count	164.3±70.7	160.1±57.4	117.6±52.5	123.6±60.0	125.6±60.1	128.2±59.7	115.2±41.8	91.13±34.1	112.1±60.1	114.9±58.6	108.6±56.9	113.3±71.0
Complications
Ascites	0	0	0	9*	10	10	11	13	26	26	18	18
Variceal bleed	0	0	0	4*	0	0	2	2	15	15	9	11
HE	0	0	0	10*	13	8	6	10	30	28	15	18
SBP	0	0	0	0	2	2	0	1	0	0	1	4
Drugs
Lactulose	0	0	0	12*	13	13	6	9	28	29	15	17
Rifaximin	0	0	0	9*	9	9	3	6	27	28	12	13
PPI	17	17	10	15*	9	10	10	10	22	24	13	12
SBP proph	0	0	0	0	2	2	0	1	1	1	1	4
Beta-blockers	20	21	11	11	7	8	4	7	18	20	6	5
Statins	18	19	6	8	4	3	4	2	9	10	1	2

^*p<0.05 Wilcoxon signed rank (proportion) or paired t-test (continuous variables), Base=baseline; data presented as raw numbers or mean±SD

Microbial analysis:

Baseline comparisons:

Differential features: There was a higher Bacteroides spp along with SCFA producers (Oscillibacter spp) with urease-negative Streptococcus spp in those who remained compensated while potential pathobionts (opportunistic pathogen that emerges due to microbial community perturbations) belonging to Enterococcus and Fusobacterium spp were higher along with Lactobacillus and Campylobacter spp in the group who developed their first decompensation (Figure 2). At baseline, those who developed second decompensation had higher Lactobacillus spp, Methanobrevibacter smithii and other pathobionts such as C.difficile. In contrast, those who remained stable at 1-decompensation had higher Lachnospiraceae, Alistipes and potentially probiotic Lactobacillus rhamnosus.(Figure 3A/B). Patients who remained stable with >1-decompensation had higher Eubacterium, Butyricoccus and Bifidobacterium spp, while their counterparts who developed even more events had higher Candida, Pseudomonas, Serratia, Veillonella spp and Lactobacillus phage Lc-Nu (Figure 3 C/D). While α-Diversity was similar, β-diversity was significantly different between compensated patients who remained stable versus not and those with 1-decompensation who remained stable versus not on PERMANOVA for all bacteria, viruses, and archaea using Biomminer (Figure 4).

Figure 2:

Baseline Microbiota in Compensated patients who remained stable versus developed first decompensation. Log2fold changes on DESEq2 of top 20 features in each subgroup at baseline. Subpart A shows the species that were higher at baseline group that remained stable, represented in blue, while Subpart B shows the species that were higher at baseline in those who ultimately experienced their first decompensation in orange ***:p<0.0001, **:p<0.001, X-axis represents log-2fold changes

Figure 3:

Baseline microbiota in groups with prior decompensation based on those who developed further outcomes Log2fold changes on DESEq2 of top 20 features at baseline. Subpart A/B: patients with 1 decompensation; A: higher in stable, B: developed second decompensation

C/D: those with >1 decompensation C: higher those who remained stable, D: higher in those who developed further decompensation ***:p<0.0001,**:p<0.001X-axis represents log-2fold changes

Figure 4:

The proportional abundance of Archaea in all samples. Comp: compensated, Decomp: decompensated,

Archaea: 18.2% of samples showed two taxonomic groups, Methanobrevibacter smithii and Methanobrevibacter unclassified(Figure 5) and those who had 1-decompensation and remained stable had higher Methanobrevibacter smithii versus those who developed their second decompensation using DESeq2.

Figure 5:

Beta-diversity measures at baseline between groups who developed outcomes versus those who remained stable. Blue: those who remained stable, Orange: those who developed first (A) or further decompensation (B and C); P value based on PERMANOVA analysis. A (compensated) and C (>1-decompensation) groups showed significant separation in microbes of those who were stable versus not at baseline, while there was a trend towards separation of microbiota at baseline for B (1-decompensation)

Longitudinal Change α-diversity:

For bacteria, we found fewer taxa in study-end samples compared to baseline in those who developed their first decompensation and those who developed further decompensations compared to baseline (Figure S1); no changes in viral α-diversity was seen. Wilcoxon tests pre vs post for bacteria showed a reduction in Chao1 in those who developed first decompensation and those with multiple decompensations from Pre to Post (Table 3). The only viral metric significantly different was a decrease in Chao1 in >1-decompensation patients using post-pre and in a comparison using paired Wilcoxon tests of Pre vs Post. β-diversity: patients with an unstable course had a greater difference in bacterial β-diversity between the 2 timepoints than those who remained stable (Compensated, p=0.007, >1-decompensation, p=0.025) and a trend for these changes in those with 1-decompensation (p=0.067). No change was seen for viruses. PERMANOVA showed a significant change compared to baseline with respect to viral and bacterial species only in unstable compensated patients (Figure 6).

Table 3:

Pre and post change in Alpha diversity using Paired Wilcoxon Tests

Group	Bacterial species	Baseline		End		p value	Viral Genera	Baseline		End		p value
		Median	IQR	Median	IQR			Median	IQR	Median	IQR
Comp stable	chao1	78	25.5	75	29	0.71	Chao1	4	3	4	3	0.87
Comp stable	Observed features	77	25.5	75	29	0.74	Observed features	4	3	4	3	0.85
Comp stable	Shannon	4.12	0.87	4.12	0.71	0.97	Shannon	0.58	0.73	0.48	0.677	0.4
Comp unstable	chao1	87	23.25	71	13	0.02	Chao1	4	3	4	2	0.58
Comp unstable	Observed features	87	22	71	13	0.01	Observed features	4	3	4	2	0.57
Comp unstable	Shannon	4.16	0.93	4.04	0.84	0.18	Shannon	0.65	0.68	0.51	0.61	0.84
1 decomp stable	chao1	72	16.25	70	20.75	0.16	Chao1	5.5	2.5	4	3	0.04
1 decomp stable	Observed features	72	16.25	69.5	20.75	0.18	Observed features	5.5	2.5	4	3	0.03
1 decomp stable	Shannon	3.59	1.06	3.66	0.92	0.87	Shannon	0.77	0.57	0.62	0.67	0.87
1 decomp unstable	chao1	83.5	19.5	72	23.5	0.37	Chao1	6	4	4	3	0.27
1 decomp unstable	Observed features	83	19.5	72	23.5	0.42	Observed features	6	4	4	3	0.37
1 decomp unstable	Shannon	3.81	0.65	3.98	0.69	0.25	Shannon	0.51	0.76	0.44	0.80	0.74
>1 decomp stable	chao1	76.5	28.5	71	27.5	0.55	Chao1	5	4.5	6	4	0.11
>1 decomp stable	Observed features	76.5	28.5	71	27.5	0.52	Observed features	5	4.5	6	3.5	0.11
>1 decomp stable	Shannon	3.26	1.28	3.48	1.14	0.45	Shannon	0.66	0.62	0.55	0.84	0.82
>1 decomp unstable	chao1	71	28	66	33.5	0.06	Chao1	8	5.75	4	4	6.00E-03
>1 decomp unstable	Observed features	71	28	66	33.5	0.06	Observed features	8	5.5	4	4	0.01
>1 decomp unstable	Shannon	3.66	1.57	3.21	1.14	0.83	Shannon	0.78	0.67	0.48	0.79	0.12

Comp: compensated, Decomp: decompensated, stable: did not develop decompensation or further decompensations, unstable: developed first or further decompensation.

Figure 6:

Longitudinal change in Beta diversity over time.

A, C and E: Bacterial species, B, D and F: Viral Genera Lighter colored dots are pre while the darker colored ones are post values. P values based on PERMANOVA Stable: those who were unchanged over time, Unstable or unst: those who developed decompensation. Data show that the greatest shift in beta-diversity over time was in viral genera and bacterial species of those who developed their first decompensation compared to their baseline.

Longitudinal change in specific bacteria and viruses and archaea(Figure 7). There was an increase in relative abundance of Faecalibacterium prausnitzii over time in compensated patients who remained stable. After first decompensation, Lactobacillus spp. and Streptococcus parasanguinis increased, while Lachnospiraceae spp. and Eubacterium hallii decreased versus baseline. There was a reduction in Myoviridae and increase in Siphoviridae viral families in patients with first decompensation versus baseline. No others were seen in the remaining groups. No differences were detected in the proportion of patients with/without Archaea by stability-status using Fisher’s Exact test; compensated Pre (p=0.36 and Post p=0.76), 1-decompensation (Pre p=0.15 and Post p=1.00) and >1-decompensation (pre p=0.70 and post p=1.00) or using the hurdle analysis.

Figure 7:

Longitudinal Changes in Specific Viral and Bacterial taxa for patients who were compensated at baseline. There was an increase in F.prausnitzii in stable patients over time. In those who developed their first decompensation, there was a reduction in Myoviridae, Lachnospiraceae spp and Eubacterium, and an increase in Lactobacillus spp and Siphoviridae. Pre: baseline, Post: after second collection, Stable: remained compensated, Unstable: developed their first decompensation

Adjustment for clinical variables: showed that only a few variables (MELD, PPI, and rifaximin) were additive to microbiota using the generalized linear model in those who remained stable or not in the group with >1-decompensations (Table 4 and Figure S2). Holdemania in >1-decompensation when accounting for rifaximin and MELD score, and Enterococcus and rifaximin and MELD score, and Sutterella and Eubacterium for PPI use and MELD were significantly affected.

Table 4:

Significant results showing within-subject different between baseline and end samples adjusted for clinical variables.

Compensated at baseline (group 1)			1 prior decompensation at baseline (Group 2)			>1 decompensation event at baseline (Group 3)
Feature	Clinical feature	P value	Feature	Clinical feature	P value	Feature	Clinical feature	P value
Enterococcus	Rifaximin	0.0002	Eubacterium eligens	PPI	0.002	Holdemania	Lactulose	0.009
Roseburia	MELD	0.009	Anaerostipes hadrus	Lactulose	0.003	Holdemania	Rifaximin	0.008
Burkholderiales bacterium 1_1_47	Lactulose	0.001	Eubacterium eligens	MELD	0.002	Holdemania	MELD	0.006

Comparison of taxa at end of decompensating events:

We compared second samples of those after their first decompensation to those post-second decompensation and those who developed further decompensations to determine the impact of the first versus further decompensating events on the gut microbiota. There were no changes in viruses. When we compared first to second decompensation, there was an over-representation of potentially autochthonous species (Oscillibacter, Subdoligranulum, Blautia, Ruminocuccus), along with urease-negative Streptococcus spp., Eubacterium spp., and Bacteroides spp and Enterococcus spp post-first decompensation but higher potential pathobionts (C.difficile),gram-negative taxa (Campylobacter, Sutterella), and Lactobacillus spp. in patients who developed a second decompensation (Figure 8). There was also a significant difference in the beta-diversity.

Figure 8:

Post-decompensation comparison between those with their first decompensation and those with their second decompensating event.

Comparison performed using DESeq2 and log2fold change data presented.

A: Top 20 species that were higher in patients after their first decompensating event are in blue X-axis represents log-2fold changes

B: Top 20 species that were higher in patients after their second decompensating event are in orange X-axis represents log-2fold changes

C: PERMANOVA of Bray-Curtis distance after the first decompensation (blue) or after the second decompensation (orange) showing significant separation

When post-first decompensating patients were compared to those with multiple decompensations, (Figure 9), higher potentially autochthonous and SCFA producers after the first decompensation and greater gram-negative (Citrobacter, Klebsiella, Campylobacter) and Lactobacillus spp in post-multiple decompensations. Again, a significant difference in beta-diversity was seen.

Figure 9:

Post-decompensation comparison between those with first decompensation and those with >2 decompensating events. Comparison performed using DESeq2 and log2fold change data presented.

A: Top 20 species that were higher in patients after their first decompensating event are in blue X-axis represents log-2fold changes

B: Top 20 species that were higher in patients after their >2 decompensating event are in orange, X-axis represents log-2fold changes

C: PERMANOVA of Bray-Curtis distance after the first decompensation (blue) or after the >2 decompensating events (orange) showing significant separation

We also compared post-first decompensation patients to baseline patients with only one decompensation and found no changes in alpha-diversity (Shannon 2.39±0.86 vs 2.35±0.49, p=0.78, Simpson 0.79±0.26 vs 0.81±0.11, p=0.55, Evenness 0.56±0.19 vs 0.56±0.11, p=0.97, Richness 69.0±17.1 vs 67.1±14.3, p=0.60 using unpaired t-tests) and beta-diversity (PERMANOVA p=0.09).

Correlation network analysis:

At baseline, those without prior decompensation who ultimately decompensated had higher clustering coefficients, centralization, and denser and homogeneous bacterial-viral-archaeal correlations compared those who remained stable (Table 5). The same pattern was seen in the remaining groups, where the clustering, centralization and density were higher at baseline for those who developed outcomes versus those who remained stable. After developing the first decompensation there was a continued higher of clustering, density, and lower heterogeneity versus baseline. On the other hand, there was a reduction in density and clustering and more heterogeneity after further decompensation in previously decompensated patients. In the groups that remained stable, there was a greater change towards clustering and density but even in the post-state, these variables were lower than those in the pre-state of those who ultimately decompensated.

Table 5:

Trans-kingdom Correlation network analyses

	Compensated and Stable		Compensated and developed first decomp
	Baseline	End	Baseline	End
Clustering coefficient	0.040	0.073	0.084	0.092
Network centralization	0.105	0.117	0.162	0.173
Avg. Number of neighbors	4.702	5.299	11.563	8.904
Number of nodes	168	154	197	188
Network density	0.028	0.035	0.059	0.048
Network heterogeneity	0.835	0.801	0.625	0.881
	1 prior decomp and stable		1 prior decomp and developed second decomp
	Baseline	End	Baseline	End
Clustering coefficient	0.078	0.087	0.132	0.080
Network centralization	0.188	0.165	0.196	0.154
Avg. Number of neighbors	11.511	10.911	11.608	9.930
Number of nodes	176	191	189	172
Network density	0.066	0.057	0.062	0.058
Network heterogeneity	0.758	0.803	0.797	0.822
	>1 prior decomp and stable		>1-decomp and developed further decomp
	Baseline	End	Baseline	End
Clustering coefficient	0.049	0.067	0.136	0.075
Network centralization	0.154	0.200	0.196	0.244
Avg. Number of neighbors	5.829	7.359	9.349	7.074
Number of nodes	199	181	175	175
Network density	0.029	0.041	0.054	0.041
Network heterogeneity	1.083	1.012	0.844	1.133

Nodes represent the individual objects/units that are being related to one another in networks. Clustering and density are measures of centralization, while heterogeneity represents the spread of the network differs between groups. These measures can be used to compare multiple networks and linkages between specific nodes. Decomp: decompensation

Archaea-bacterial linkages were only seen in compensated patients and those with their second decompensation (Figure 10). Methanobrevibacter linkages were negative with Clostridium spp., which was relatively stable in those who remained compensated at both sample collections. However, in those who developed their first decompensation, the more complex correlation between Archaeal species and bacteria reduced post-decompensation with new negative linkages with Veillonella spp. In Groups 2 and 3 there was again a reduction in archaeal-bacterial complexity after further decompensation, which was again centered around Veillonella spp. before and after (Figure S3) with negative linkages between archaea and Veillonella spp. Viral-bacterial subnetworks were centered around C2like viruses in all subgroups. Negative linkages were seen between C2like viruses with potential autochthonous taxa at baseline as well as with C.leptum after the first decompensation (Fig 10). Similar negative linkages of relatively beneficial taxa belonging to Faecalibacterium and Lachnospiraceae spp with C2Like viruses and Lactobacillus phage Nu was found in groups 2 and 3, which persisted post decompensation (Figure S4).

Figure 10:

Correlation networks centered on Archaea and Viruses in Compensated Patients across both timepoints.

A: Archaea-bacterial subnetwork for stable patients at baseline, B: Archaea-bacterial subnetwork for stable patients at study end, C: Archaea-bacterial for unstable patients at baseline, D: Archaea-bacterial subnetwork after the first decompensation, E: Viral-bacterial subnetwork for unstable patients at baseline, F: Viral-bacterial subnetwork after the first decompensation. No viral-bacterial correlation was seen for stable patients. Pink: bacteria, Green: viruses, Teal: Archaea, red lines: positive correlation, blue lines: negative correlation. Pre: baseline, Post: post first decompensating event, stable: remained compensated, unstable: developed first decompensation

DISCUSSION:

Our study results demonstrate longitudinal trans-kingdom gut microbial changes in cirrhosis are more marked in those who develop their first decompensating event compared to those developing further decompensation. At baseline there are differences in predominantly bacterial composition between those who ultimately decompensate versus those who remain stable. After decompensation, there is reduction in bacterial alpha-diversity and change in beta-diversity compared to baseline, along with a relative reduction in commensal organisms and an increase in pathobiont abundance. While stand-alone viral and archaeal changes were not markedly different apart from higher Lactobacillus phage Lc nu in those who ultimately developed their first decompensation, trans-kingdom correlation networks are more cohesive, homogenous, and clustered at baseline in those who develop first, or further decompensation compared to those who remained stable.

Development of the first decompensation is a critical milestone in cirrhosis progression and understanding the pathophysiology of this progression could create newer therapeutic and prognostic targets[10, 21]. To date, most research in the cirrhosis-related microbiome, including from our group, has occurred using one sample and projection of clinical outcomes that occur at varying timepoints weeks to years after that based on that assessment[13, 22, 23, 24]. This study extends these by studying the same subjects twice in those who remained stable and those who developed a clinically relevant decompensating event. This novel study design allows us to understand not only the predictive potential of microbiota while using subjects as their own controls, but also defines impact of decompensation on the bacterial, viral, and archaeal microbial structure and interactions.

We found that most changes occurred in the bacterial kingdom, predominantly in those who developed outcomes compared to their pre-outcome baseline. The most pronounced impact was in subgroups who were compensated at baseline who developed their first decompensation, with relatively lower α/β-diversity measures, along with reduction in commensals and SCFA producers and higher pathobionts. Specific taxa associated with beneficial metabolites such as Eubacterium, Oscillibacter and Lachnospiraceae are over-represented in patients who remain stable or compensated while potential pathobionts were higher even at baseline in those who ultimately developed their first or subsequent decompensating events[22, 25].

These differences persisted and were only marginally different when important clinical variables such as rifaximin and MELD were included, indicating a shift in the gut microbiome as a whole because of the decompensation rather than an epiphenomenon of medications[26, 27]. To add to this, the beneficial species Faecalibacterium prausnitzii was higher over time in compensated patients who remained stable[27]. Moreover, the post-decompensation comparison between those with their first decompensation versus subsequent ones showed that the number of events could be linked with greater loss of SCFA-producing bacterial relative abundance. This was accompanied by higher Lactobacillus spp. and pathobionts in those with multiple decompensations compared to first decompensation. Importantly, there remained significant microbial differences in those with 2 versus >2 decompensations. This novel finding indicates that even within patients, the number of prior decompensating events matters and not just whether they have decompensated in the past. This is in line with current BAVENO VII consensus where first decompensation is considered a major sentinel event that may be distinct from subsequent decompensations[3]. As we move towards potentially using the gut microbiome for individualizing therapeutic and prognostic approaches in cirrhosis course, we need to adjust for the number and trajectory of the decompensating events. These data could guide clinicians, researchers, and trialists to potentially use gut microbiome composition as a means to predict outcomes and use microbially-targeted therapied focused on individual patient decompensation history.

Interestingly, viral changes were not as marked but there were changes in two families of the phage order Caudovirales, including a reduction in Myoviridae and an increase in Siphoviridae after first decompensation. In correlation network analyses, Siphovirideae constituents, ceduoviruses or C2-like phages were negatively linked with SCFA-producers at baseline and again negatively associated with C.leptum after the first decompensation. Siphoviridae are lytic phages and C2-like viruses, especially, target lactate-producing bacteria, which are usually increased in cirrhosis with poor outcomes[28, 29]. One specific phage within Siphoviridae that was in higher abundance at baseline in those who decompensated further was Lactobacillus phage Lc nu which is lytic against beneficial L.rhamnosus strains[30]. The negative correlation of this phage with beneficial taxa even in those who had developed decompensation could also signify the potential to engender negative outcomes. Myoviridae have been associated with protection from poor outcomes, especially HE and hospitalizations in our prior cross-sectional study but this validates that prediction using longitudinal analyses[13].

We also found that Archaea were found in appreciable abundance only in about a fifth of patients and did not show a trend towards a subgroup or timepoint apart from an increase at baseline in those who developed their second decompensation. The proportion is similar to prior studies[31]. The abundance of archaea vis-à-vis lactulose and rifaximin is important since methane can potentially slow down intestinal transit[32]. Correlation network analysis for archaea showed a pronounced reduction in bacterial-archaeal linkage after the first decompensation while these remained similar across stable patients. The primary linkages between Methanobrevibacter spp. were with Clostridia, Eubacterium and Veillonella spp. Veillonella is a lactate consumer which then produces propionate, while other lactate consumers such as Eubacterium hallii and Eubacterium rectale, and Flavonifactor spp. produce butyrate after consumption of lactate[33, 34]. Unchecked lactate production, which could occur with lactulose and advancing cirrhosis without adequate bacteria to consume it and convert it into SCFA can potentially lead to impaired intestinal integrity and potentially brain dysfunction[35, 36]. Moreover, in ruminant studies, higher lactate is associated with lower archaeal methane production via multiple mechanisms[37, 38]. Intestinal methane is associated with slow intestinal transit while lactulose use is intended to reduce this[38]. This could also explain the negative associations with lactate consumers and butyrate producers with Methanobrevibacter spp. The overall impact of this interaction is unclear but the functional implications of lactate, SCFA, and methane could be important in modulating therapeutic approaches. These differential interactions could also explain the greater cohesion of the trans-kingdom correlation network in those who ultimately decompensated versus those who remained stable, which has been shown in studies of infected patients with cirrhosis[9]. In addition, the relative stability of bacteria-Archaea interaction in stable patients over both timepoints versus a collapse in those who were post-outcomes, also suggests an important role of non-bacterial components of the microbiome in cirrhosis.

An important consequence of decompensation is the initiation, modification, or removal of therapies that could in turn impact the microbiota[10, 39]. To ensure that the patients had reached steady state post-decompensation, we only included outpatients at least one month after the event. While more recent studies have used a 3-month window, we used this timepoint ensure the shortest time to a stable clinical situation since several patients develop multiple episodes of decompensation once this cycle is initiated[40]. About half of these patients were started on lactulose and/or rifaximin, while almost a quarter were started on PPIs. Statin and beta-blocker use remained largely similar. GLM analyses showed higher Enterococcus associated with rifaximin use, which likely reflects the worsening liver disease, since a similar pattern was seen with MELD score change [41]. Similarly, Holdemania and rifaximin association in >1-decompensated patients and Eubacterium eligens in 1-decompensation patients affected by PPI, were both similar in pattern of change as the MELD score. None of the other medications, statins, or beta-blockers were additive. Ultimately, the changes noted here with medications on composition of bacteria were not likely contributory to the changes over and above decompensation since a majority of these were linked with the advancement in liver disease itself.

These distinctions are important as more clinical evidence is being published about the unique routes patients with cirrhosis take towards decompensation[3]. Therefore, to use microbiota as potential biomarkers, the impact of pre- and post-medications, disease severity, and type of decompensation is important to clarify. These data have the potential to prognosticate patients, develop microbiome-specific tools to potentially reduce these events, and determine candidacy for microbiome-related therapies such a pre/pro and post-biotics and fecal microbiota transplants[6]. Patients with multiple episodes of decompensation have a differing microbial trajectory compared to those after their first decompensation. Defining the bacteria, archaea, and viruses in patients with cirrhosis would help individualize prognostic strategies and refine eligibility and selection for success for fecal microbiota transplant and treatments for HE and SBP. In addition, knowledge of the interactions of bacteria with phages could help define therapies to protect against decompensation and development of infections[9].

Our study is limited by the relatively modest numbers, but the pre and post design of patients who remained stable versus not helps reduce the inter-individual variability. Most patients developed HE as their decompensating event, which requires the use of microbially-acting medications[15]. This is reflective now of the increasing HE burden but these data are unlikely to be due to HE-medications due to the relatively low impact on microbial composition and the correlation of these medication-related changes with MELD score[42, 43, 44, 45]. For viral analysis, did not enrich for viral-like particles and our sequencing strategy only captured DNA viruses, and excluded RNA viruses, however the conventional VLP concentration followed by sequencing could also bias isolation of viral taxa [12, 46, 47]. We did not focus on fungi using ITS sequencing given their relative sparseness but since shotgun metagenomics annotates some eukaryotic DNA as well, we did find some non-significant mycobiota trends[48]. Lastly, we did not study the functional or immune aspects of the gut microbiome[11, 49]. We conclude that longitudinal sampling of patients with cirrhosis shows unique microbial signatures of first versus subsequent decompensating events. Interactions between lactate producers and consumers, methanogens and short-chain fatty acid producing taxa could modulate these microbial changes and their impact on the hosts. The change in microbial structure is most pronounced after the first decompensating event compared to subsequent ones. Archaea and viral contributions to the overall complications of cirrhosis are modest compared to bacterial inputs but viral-bacterial-archaeal correlation network density and clustering at baseline can differentiate between patients who develop future decompensating events compared to those who remained stable. These include Methanobacter brevii, C2-like viruses and Lactobacillus phage Lc nu which could modulate outcomes in patients before and after decompensation. Trans-kingdom changes in gut microbiome could have important implications in decompensation and further decompensation in patients with cirrhosis.

What is already known on this topic

• Development of first and subsequent decompensations are critical milestones in the natural history of cirrhosis

• Cirrhosis is associated with altered gut microbiota, which change with decompensating events, but longitudinal studies across various stages of decompensation need to be performed

• Methanogens, such as Archaea are associated with a slower intestinal transit that could potentiate hepatic encephalopathy (HE) development, but their role in cirrhosis is unclear.

What this study adds

• In this longitudinal study of patients who developed their first and subsequent decompensation compared to demographically and cirrhosis-severity balanced patients who remained stable, the changes in bacteria, viral and archaea are greatest in those experiencing their first decompensation.

• Commensal bacteria are lower and pathobionts are higher at baseline in those who subsequently develop their first or subsequent decompensation.

• Archaea are present in a fifth of individuals but are not related to cirrhosis severity, however, correlations between archaea, viruses, and bacteria are strongest in those who ultimately decompensated compared to those who remained stable.

• Lactobacillus phage and C2-like viruses from Siphoviridae associate with decompensation by linking negatively against probiotic and commensal taxa.

• Number of decompensating events affect microbiota differently irrespective of medication use, therefore the individual patient trajectory towards decompensation, is important for appropriate microbial interpretation and prognostication.

How this study might affect research, practice, or policy

• Profiling of gut bacteria, viruses, and archaea in compensated patients could potentially prevent the development of the all-important first decompensation through better prognostication.

• These data could enhance precision targeting of microbiota to influence interactions between methanogens, lactate-producers and consumers, and SCFA producers by potentially using specific viruses and lytic phages such as Lactobacillus Lc nu.

• The knowledge of non-bacterial components of the gut microbiome such as archaea and viruses in prediction of outcomes and change with time could improve selection of patients for microbially-based therapies and improve overall prognostication to prevent future decompensation.