Elderly patients have an altered gut-brain axis regardless of the presence of cirrhosis

Cognitive difficulties manifested by the growing elderly population with cirrhosis could be amnestic (memory-related) or non-amnestic (memory-unrelated). The underlying neuro-biological and gut-brain changes are unclear in this population. We aimed to define gut-brain axis alterations in elderly cirrhotics compared to non-cirrhotic individuals based on presence of cirrhosis and on neuropsychological performance. Age-matched outpatients with/without cirrhosis underwent cognitive testing (amnestic/non-amnestic domains), quality of life (HRQOL), multi-modal MRI (fMRI go/no-go task, volumetry and MR spectroscopy), blood (inflammatory cytokines) and stool collection (for microbiota). Groups were studied based on cirrhosis/not and also based on neuropsychological performance (amnestic-type, amnestic/non-amnestic-type and unimpaired). Cirrhotics were impaired on non-amnestic and selected amnestic tests, HRQOL and systemic inflammation compared to non-cirrhotics. Cirrhotics demonstrated significant changes on MR spectroscopy but not on fMRI or volumetry. Correlation networks showed that Lactobacillales members were positively while Enterobacteriaceae and Porphyromonadaceae were negatively linked with cognition. Using the neuropsychological classification amnestic/non-amnestic-type individuals were majority cirrhosis and had worse HRQOL, higher inflammation and decreased autochthonous taxa relative abundance compared to the rest. This classification also predicted fMRI, MR spectroscopy and volumetry changes between groups. We conclude that gut-brain axis alterations may be associated with the type of neurobehavioral decline or inflamm-aging in elderly cirrhotic subjects.

The rapidly aging population, with the accompanying neuro-cognitive sequelae such as dementia, is a major medical and psychosocial burden 1 . This maturing of the population has also affected patients with cirrhosis, who, in addition to the cognitive issues related to aging, are also prone to hepatic encephalopathy (HE) 2 . Elderly subjects can have issues with amnestic (memory-related) and non-amnestic (unrelated to memory) cognitive dysfunction, which may be reflected in geriatric cirrhotic patients as well. The subclinical phase, known as covert HE (CHE) usually affects non-amnestic domains such as attention, visuo-motor coordination and executive function 3 . This has been shown in prior studies using mixed tests of memory and attention in which younger cirrhotics are more likely to be impaired on attention and visuo-spatial domains rather than amnestic domains such as delayed memory 4 . Further characterization of the interaction between dementia (mostly associated with amnestic issues) and HE in elderly cirrhotics is critical because an increasing number of these patients are being evaluated for liver transplant 5 . Patients with HE, unlike those with dementia, can expect a reasonable improvement in cognition after transplant and are considered appropriate for listing 6  taxa such as Ruminococcaceae and Lachnospiraceae were linked with memory-unrelated test performance. In contrast potentially pathogenic taxa such as Enterobacteriaceae were negatively linked with memory-unrelated performance (Fig. 3A,B). In the non-cirrhosis group, the linkage pattern between cognition, autochthonous and potentially pathogenic taxa was similar to cirrhotic group. In addition, LEfSe-discriminated taxa associated with the non-cirrhosis group in comparison with cirrhotics, Synergisticaeae and Peptococcaceae, were associated with poor cognitive performance (negatively associated with HVLT score).

Discussion
This study demonstrates that elderly cirrhotic patients have a significantly impaired cognitive performance and HRQOL, associated with systemic inflammation, gut dysbiosis and altered MR spectroscopic findings compared to age-matched non-cirrhotic subjects. Furthermore, this altered gut-liver-brain axis, especially from a multi-modal MRI perspective, is refined when neuropsychological profile rather than cirrhosis itself is used as a classifier. Changes in systemic inflammation and gut microbiota that track the neuropsychological classification suggest a potential biological basis for this division that transcends the diagnosis of cirrhosis. Similarly, neuropsychological impairment was found to be the earliest biomarker for predicting progression of mild cognitive impairment and Alzheimer's disease 13,14 .
The growing elderly cirrhotic population, who could have concomitant amnestic and non-amnestic cognitive deficits, remains an important concern for clinicians and neuropsychologists. The converging effects of CHE and age-related changes suggestive of mild cognitive impairment in these patients could impact the HRQOL and potentially disease progression. Our findings demonstrated that the neurocognitive profile of elderly cirrhotic patients was largely similar to what could be expected in younger cirrhotics compared to non-cirrhotic individuals with a few exceptions 3 . Elderly cirrhotic subjects had impairment on selected memory-related tests that would not be expected in younger cirrhotic patients 3 . The relationship between cirrhosis and aging was therefore further explored using the neuropsychological classification. This classification showed that almost half of the patients classified as amnestic type and more than three quarters of amnestic/non-amnestic type had underlying cirrhosis. This higher burden of cognitive impairment in cirrhotic patients, regardless of classification modality, translated into a worse HRQOL that spanned most areas of daily functioning. Therefore elderly cirrhotic patients, even without prior overt HE, are likely to have greater amnestic impairment, which translates into a poor HRQOL compared to non-cirrhotic individuals.
While the relatively simple classification according to underlying cirrhosis was important to define HRQOL changes, it did not readily translate into corresponding changes in fMRI or volumetric analyses. On the other hand the neuropsychological classification determined that patients with amnestic/non-amnestic impairment required greater neuronal recruitment of the visuo-spatial network to achieve the same response compared those with amnestic impairment alone, or unimpaired subjects regardless of cirrhosis 15 . A similar pattern was observed on volumetry where cirrhosis vs. no-cirrhosis did not significantly differ while the neuropsychological classification clearly separated the groups. Interestingly, hippocampal and brain volume decrease was found in those with predominant amnestic type. A lower brain volume has been associated with non-cirrhotic amnestic impairment and our study found similar trends in patients with amnestic impairment, which included cirrhotic patients 13 . The relative higher brain volumes in the predominantly cirrhotic amnestic/non-amnestic group likely reflect cirrhosis-associated brain edema 16 . These findings were further extended by the MR spectroscopy data. This showed that elderly cirrhotic patients, in addition to demonstrating ammonia-associated changes of higher creatine-adjusted values of Glx and lower mI, and had a lower creatine-adjusted NAA + NAAG, which is a neuronal marker described again in non-cirrhotic amnestic impairment 17 . This MR spectroscopic classification was further refined by the cognitive classification that reaffirmed the lower creatine ratios of NAA + NAAG and mI in the most impaired groups. Therefore, from a multi-modal MRI perspective, where changes are often subtle and may be subclinical, a division based on neuropsychological performance may help define brain dysfunction better than a simplistic cirrhosis/no-cirrhosis classification.
However, the pathophysiology of this cognitive impairment could span several other organ systems, given the complex interaction between systemic inflammation, gut microbial dysbiosis and brain function in younger cirrhotic patients 8,18,19 . When taken as a whole group, patients with cirrhosis regardless of neuropsychological performance had higher relative abundances of Lactobacillales members. These taxa were positively linked with good performance on memory-based tests, which could potentially explain the lack of major differences on these tests between the cirrhotic and non-cirrhotic groups. However, the non-cirrhotic group had higher relative abundances of Synergistaceae and Peptococcaceae, which were related to poor amnestic performance. Prior known potentially beneficial, autochthonous families such as Ruminococcaceae and Lachnospiraceae were positively linked, while potentially pathogenic taxa such as Enterobacteriaceae were negatively linked to good cognition regardless of cirrhosis 19 .
Intriguingly changes in gut microbiota persisted when groups were classified neuropsychologically. Fecalibacterium and Butyricoccus, which are potentially probiotic genera associated with inflammation suppression and butyrate production respectively, were higher in cognitively unimpaired subjects [20][21][22] . On the other hand, genera belonging to beneficial, autochthonous families were less likely in patients with amnestic/non-amnestic type compared to those with amnestic type alone 19 . Interestingly, Porphromomadaceae, has been associated with neuronal dysfunction in younger cirrhotics, were more likely to be present in amnestic patients and was associated with poor cognition in the cirrhosis correlation network 9,11 , alluding to the predominant neuronal role in the progression to dementia. Interestingly, unlike in younger cirrhosis vs. no-cirrhosis comparisons, we did not find members of Proteobacteria phylum as differentiators. This could reflect prior studies that show that elderly subjects have a baseline higher Proteobacteria abundance overall, and could also explain the relatively insignificant overall changes in endotoxin 23 . The other inflammatory markers were chosen to represent innate immunity Scientific RepoRts | 6:38481 | DOI: 10.1038/srep38481 Unimpaired (n = 23) Amnestic type (n = 25) Amnestic/non-amnestic type (n = 28) P value (IFN-γ ), prostaglandin pathway (PGE2) and systemic inflammation (IL-6). PGE2 was analyzed because prior non-cirrhotic human and animal reports have alluded to it as a marker of pre-clinical dementia, but it was not able to differentiate the current population 24 . A higher IL-6 is often seen in cirrhotic patients, which could explain the higher values in the amnestic/non-amnestic group 25 . A higher IFN-γ in both cognitively impaired groups could reflect a non-specific innate response to the bacterial dysbiosis 26 .
This differential systemic milieu could link microbial and brain changes but endotoxemia does not seem to play an important role 27 . However, further mechanistic studies are needed. Beneficial changes in butyrate and inflammation have been shown with use of synbiotics in elderly patients, noting that these factors may be another aspect to improve the impaired inflammatory-immune basis of inflamm-aging 8,12,28 . Further studies focusing on improving the altered gut-brain axis in cirrhotic elderly patients and potentially using specific stool bacteria as biomarkers for disease progression are needed.
We excluded OHE patients, who by definition have worse cognition, brain MRI results and gut dysbiosis compared to others, because it could have influenced the interpretation of cognitive tests, and gut microbiota analysis 19,27 . However, we were still able to find differences in the population with subclinical cognitive dysfunction, and would expect these changes to be magnified in future studies with elderly OHE patients. Another limitation of this study is that it represents a highly selected subject sample without significant co-morbid conditions. The relative contribution of cirrhosis, dysbiosis and aging on the ultimate cognitive and brain MR function is difficult to delineate given the sample size, and larger studies are required.
Our results suggest that an altered gut-brain axis occurs in elderly patients with differential changes based on the underlying cirrhosis and the type of cognitive dysfunction. Elderly cirrhotic patients are likely to have significant amnestic or memory-related impairment, which is usually associated with dementia and translates into a  Table 3. Activation table of brain areas showing group differences during correct inhibition to lures; no differences were found between cirrhosis vs. no-cirrhosis and between amnestic group and unimpaired patients. a Presents the brain areas where the brain activation to correct inhibition to lures was significantly greater in the Amnestic/Non-Amnestic compared to Amnestic type groups on fMRI. b Presents the brain areas where the brain activation to correct inhibition to lures was significantly greater in the Amnestic/Non-Amnestic compared to the Unimpaired type group on fMRI. A 'Cluster' is a group of anatomically adjacent voxels found to be statistically significant after thresholding with z-threshold = 2.3 and p = 0.05. Z-score represents the magnitude of group differences in activation and Montreal Neurological Institute (MNI) coordinates (x, y and z axis) represent a standardized way of presenting the anatomical location of the peak Z-score on a normalized brain template.
poor HRQOL. A classification based on neuropsychological performance further refines changes in multi-modal MRI markers of functional and structural cortical connectivity. Further studies are needed in more advanced patients to define the effect of these changes on outcomes, and to utilize the gut microbiota as an alternative approach to diagnose and treat cognitive disorders in the growing elderly cirrhotic population.

Methods
We enrolled consecutive outpatients between the 65 and 85 years of age after informed consent. We excluded patients with significant physical (congestive heart failure, chronic obstructive pulmonary disease, HIV, malignancies, dialysis-dependent, diabetes with HbA1c > 7%) and neuropsychiatric conditions (established dementia, prior stroke, bipolar disorder, schizophrenia, Parkinson's disease, multiple sclerosis, seizures). We also excluded all on anti-psychotic, anti-seizure, benzodiazepines and chronic anti-depressants other than selective serotonin reuptake inhibitors (SSRI). We did not exclude patients on chronic SSRI given the drugs' minimal effect on cognitive performance in cirrhosis 29 . We also excluded subjects with alcohol misuse or illicit drug use within 3 months,

Figure 1. Group difference contrasts on brain activation on functional MRI during correct inhibition to lures on the inhibitory control tests using a mixed-effects analysis.
The cluster-forming threshold was created with z = 2.3 with a corrected p < 0.05. The red-yellow schema is based on levels of gradation of differences from 2.3 to > 4 between groups. (A) Comparison of amnestic/non-amnestic type to amnestic type patients in which amnestic/non-amnestic type patients required a greater brain activation extent to achieve the similar response on lure inhibition compared to amnestic type patients. (B) Comparison of amnestic/non-amnestic type to unimpaired patients in which amnestic/non-amnestic type patients required a greater brain activation extent to achieve the similar response on lure inhibition compared to unimpaired subjects.  Table 4. Group differences in brain metabolite creatine ratios in the anterior cingulate cortex. ‡ p < 0.05, ‡ ‡ p < 0.01 between cirrhotic and non-cirrhotic, *p < 0.05, **p < 0.01. Unimpaired vs. amnestic + non-amnestic type; † p < 0.05, † † p < 0.01 amnestic vs. amnestic + non− amnestic type. and those who had used laxatives, probiotics and antibiotics within the last six weeks. Cirrhotic patients were excluded if they had current/prior overt hepatic encephalopathy (OHE) 19,27 . After enrollment, all subjects underwent mini-mental status testing (MMSE) and the Beck Depression Inventory-II (BDI-II). A score < 25 on MMSE and > 12 on BDI-II excluded them from further study 30 . Demographic information and severity of cirrhosis were Thalamus L 6.75 ± 0.5* 6.09 ± 0.7 † † 7.05 ± 0.6 Thalamus R 6.53 ± 0.5 6.18 ± 0.4 † † 6.81 ± 0.5 Table 5. Group differences on brain volumetric analysis. *p < 0.05, **p < 0.01. Unimpaired vs. amnestic type; † p < 0.05, † † p < 0.01 amnestic vs. amnestic/non-amnestic type. For total brain volumes (white matter, gray matter and total brain) data are presented as x 10 −5 mm 3 while for local volumes (Hippocampus and Thalamus) data are presented as x 10 −3 mm 3 . All subjects' volumes are reported relative to a normalized skull size. The cladogram shows the phylogenetic relationship between the bacterial families that were higher in non-cirrhotic (green) compared to cirrhotic patients that are represented in the red. (B) LEFse comparison between amnestic/non-amnestic type and amnestic type patients. The cladogram shows the phylogenetic relationship between the bacterial taxa that were higher in amnestic/non-amnestic type (green) compared to amnestic type patients that are represented in the red. (C) LEfSe comparison between unimpaired and amnestic/non-amnestic type patients. The cladogram shows the phylogenetic relationship between the bacterial taxa that were higher in unimpaired patients (red) compared to amnestic/non-amnestic patients. (D) LEfSe comparison between unimpaired and amnestic patients. The cladogram shows the phylogenetic relationship between the bacterial taxa that were higher in unimpaired patients (green) compared to amnestic patients (red).  and Streptococcaceae, all of which belong to Lactobacillales and are higher in cirrhotic compared to noncirrhotic individuals, were associated with good cognitive performance. As expected most cognitive tests were linked with each other in the expected directions. (B) Correlation network for non-cirrhotic elderly subjects. Significant positive correlations were seen between Ruminococcaceae and good cognition and negative ones between Enterobacteriaceae and cognitive performance. Peptococcaceae and Synergistaceae, which were higher in non-cirrhotics compared to cirrhotics, were actually associated with a poor cognitive performance. The correlations of Porphyromonadaceae were largely associated with poor cognitive performance.  33 and Similarities subtests 34 ]. The RBANS 33 assesses 5 cognitive domains (immediate and delayed memory, visuo-spatial skills, attention, and language). While it was originally developed for screening of dementia, it has also been used in HE, with a predominant "subcortical" pattern of dysfunction 33,4 . HRQOL assessment was performed using the Sickness Impact Profile (SIP) and the computerized adaptive PROMIS tools 35 . We collected serum for inflammatory cytokines, and stool for microbiota analysis using published methods 36 .
Inflammatory cytokines analyzed were IL-6, Interferon-γ (IFN-γ ) and prostaglandin E2(PGE2) 24 using ELISA and endotoxin was assessed using LAL assay (Assaygate, Ijamsville, MD) 19 . Stool microbiota composition was determined using published multi-tagged sequencing techniques 37 and analyzed using LEFSe (Linear Discriminant Analysis Effect Size) between the study groups 38 . Correlation network analysis between cognitive testing results, microbiota and inflammatory cytokines was performed in the cirrhotic group and in the non-cirrhotic group using a customized R package 11 . Only correlations that were p < 0.01 and r > 0.6 or < − 0.6 were studied.
Patients then underwent multi-modal brain MRI on the same day as these procedures. The MRI consisted of (1) fMRI, (2) MR spectroscopy, and (3) volumetric analysis MR Imaging Methods and analyses (Supplementary Information). All images were acquired on a Philips Ingenia 3 T scanner with a 32-channel receive head coil (Philips Medical Systems, Best, Netherlands).
fMRI ICT task. During this task the subject was asked to respond to targets and withhold responses to lures, on inhibitory control test, a validated go/no-go task, while in the scanner 39 . A brief training session was performed prior to the MRI scan. Each subject underwent six runs of the ICT inside the scanner (supplementary information).
fMRI data analysis. fMRI data analysis was carried out using the FMRI Expert Analysis Tool v 5.98 part of FSL (FMRIB's Software Library) 40 . After standard preprocessing, a time-series statistical analysis was carried out with local autocorrelation correction using the General Linear Model. The model included correct response to target and correct inhibition to lures as contrast of interest. Incorrect response to lure, random responses and six motion parameters were added as contrasts of no interest. Group average statistical maps were created in Montreal Neurological Institute (MNI) standard space for all correct inhibition to lures trials. The MNI template is an internationally recognize standard for anatomical localization of brain activation observed on fMRI. Also, statistical maps to evaluate group differences were generated and thresholded using cluster-based thresholding.
Spectroscopy analysis. The choline, creatine, myo-inositol, N-acetylaspartate + N-acetylaspartate glutamate (NAAA + NAAG) and glutamate + glutamine (Glx) complex peak areas were computed using a quantitative assessment of the metabolite concentration by means of LCModel software with the creatine concentration ratio 41 . These metabolites were chosen based on prior HE research 16 .
Volumetric analysis. Brain tissue volume, normalized for subject head size, was estimated with SIENAX 42 . Additionally, we also estimated hippocampal and thalamus volumes using part of FSL 43 and normalized them for head size.
Study Approval and Consents. All participants gave informed consent. The study is approved by the Institutional Review Boards at Virginia Commonwealth University and McGuire VA Medical Center and all experiments were performed in accordance with relevant guidelines and regulations.
Analysis of the subjects was performed using two major classifications: 1. Cirrhotic vs. non-cirrhotic patients and 2. Subjects without impairment vs. those with predominant amnestic impairment vs. those with both amnestic and non-amnestic impairment regardless of cirrhosis.
In arriving at the classification the neuropsychologist reviewed, and applied age correction, to the subject's raw cognitive scores and conducted a standardized interview with each subject and their companion, if available. During the interview academic, vocational, and substance abuse history was obtained. Subjects whose cognitive performance on all the measures ranged from -2 standard deviations and above were considered unimpaired. Subjects with focal impairment ≥ 2 SD on ≥ 2 learning/memory subtests were classified as amnestic type. Finally, subjects were considered amnestic/non-amnestic type if they were impaired ≥ 2 SD on measures in > 2 domains, such as information/psychomotor speed and learning/memory. The interview notes and cognitive performance were then reviewed by another psychologist (KS) who was blinded to the original division by the neuropsychologist. A Kappa score between the divisions was calculated for concordance. Statistical analysis. All statistical analyses were performed using the SPSS (version 12, SPSS Inc., Chicago, IL, USA) with appropriate tests. Laboratory tests, neuropsychological tests, HRQOL tests, volumetrics and MRS were compared between groups by the two-tailed unpaired t-test or ANOVA based on the divisions. Multiple comparison correction was done wherever applicable using Holm-Bonferroni method. Specialized analyses (microbiota, fMRI, correlation networks are mentioned in individual sections). A sample size of at least 16 in each group was deemed to be appropriate with an 80% power based on our prior studies on gut-brain analysis in younger patients with cirrhosis compared to healthy controls 9 .