**Cognitive Outcomes and Relationships with Phenylalanine in Phenylketonuria: A Comparison between Italian and English Adult Samples**

**Cristina Romani 1, \* , Filippo Manti 2 , Francesca Nardecchia 2 , Federica Valentini 3 , Nicoletta Fallarino 3 , Claudia Carducci 4 , Sabrina De Leo 5 , Anita MacDonald 6 , Liana Palermo 7 and Vincenzo Leuzzi 2**


Received: 26 August 2020; Accepted: 24 September 2020; Published: 3 October 2020

**Abstract:** We aimed to assess if the same cognitive batteries can be used cross-nationally to monitor the effect of Phenylketonuria (PKU). We assessed whether a battery, previously used with English adults with PKU (AwPKU), was also sensitive to impairments in Italian AwPKU. From our original battery, we selected a number of tasks that comprehensively assessed visual attention, visuo-motor coordination, executive functions (particularly, reasoning, planning, and monitoring), sustained attention, and verbal and visual memory and learning. When verbal stimuli/or responses were involved, stimuli were closely matched between the two languages for psycholinguistic variables. We administered the tasks to 19 Italian AwPKU and 19 Italian matched controls and compared results from with 19 English AwPKU and 19 English matched controls selected from a previously tested cohort. Participant election was blind to cognitive performance and metabolic control, but participants were closely matched for age and education. The Italian AwPKU group had slightly worse metabolic control but showed levels of performance and patterns of impairment similar to the English AwPKU group. The Italian results also showed extensive correlations between adult cognitive measures and metabolic measures across the life span, both in terms of Phenylalanine (Phe) levels and Phe fluctuations, replicating previous results in English. These results suggest that batteries with the same and/or matched tasks can be used to assess cognitive outcomes across countries allowing results to be compared and accrued. Future studies should explore potential differences in metabolic control across countries to understand what variables make metabolic control easier to achieve.

**Keywords:** PKU; cognitive outcomes; cross-cultural; cross-countries; Phe associations

#### **1. Introduction**

Phenylketonuria (PKU) is an inherited metabolic disease occurring in about 1/10,000 live births where an error in the gene coding for the enzyme phenylalanine hydroxylase (PAH) produces an inability to metabolize the amino acid phenylalanine (Phe) into tyrosine with serious consequences for brain health [1]. Tremendous advances have been made in our understanding and treatment of this disorder from the 1930s when it was first discovered by a Norwegian physician, Asbjørn Følling, who noticed high levels of phenylpyruvic acid in the urine of some patients with severe mental disability and established connections with high levels of Phe in the blood. Since the wide-spread introduction of new-born screening in the late seventies in most countries, infants with PKU follow a Phe-restricted diet which lowers blood Phe levels and eliminates mental disability. It is now recommended that a PKU diet is followed throughout life. Current European guidelines recommend Phe to be kept within the target range of 120–360 µmol/L till 12 years of age and within 120–600 µmol/L, above 12 years [2,3]. American guidelines recommend a target range of 120–360 µmol/L throughout life (American College of Medical Genetics and Genomics, ACMG) [4]. In classical PKU, without treatment, Phe could exceed >2000 µmol/L.

Maintaining a Phe-restricted diet allows people with PKU to lead normal lives. However, not all is well. The PKU diet is expensive, unpalatable, and unsociable. Thus, it is often self-relaxed in late childhood and abandoned during late adolescence [5–9]. Possibly because Phe levels remain suboptimal, on average, people with PKU do not reach their full cognitive potentials [7]. IQ is in the normal range, but lower than matched controls [10], and there are impairments in cognitive tasks, especially when speed of processing and higher cognitive functions are involved [11–16]. Moreover, this is the first generation of early treated adults with PKU (from now on AwPKU) to reach middle-age and we do not know the effects of prolonged high levels of Phe on aging brains.

Better management of PKU may be achieved with the wider use of existing pharmacological treatments [3,17,18] and the introduction of new ones [19–21], but it also depends on a better understanding of how the cognitive impairments experienced by this population relate to levels and variations in Phe levels across the life span. While Phe may be particularly toxic for developing brains, we need evidence of the safety of accumulating high Phe levels on aging brains. Finally, it is important to understand whether some individuals are less affected by high Phe since there is some evidence of individual variation [6,22–24]. For a few people with PKU, it may be less important to keep on a strict diet.

Establishing the efficacy of new treatments and the safety of existing ones relies on the comprehensive cognitive assessments of large samples of people with PKU and relative controls. Ideally, cognition should be tested across the board because the effects of Phe may vary for different cognitive functions. Moreover, it is important to use multiple measures to increase reliability since tasks do not tap cognitive function in a simple, univocal way. However, this is time-consuming and recruiting participants is challenging since PKU is a rare disease. These difficulties, compounded by resource limitations, mean that it is difficult for studies based on single clinical centres to achieve the desired breadth and depth of testing with enough power [15,16,25] making the ability to collate results across national and international centres particularly important. However, there is a lack of studies comparing results across national samples.

Similar negative effects of PKU on cognition have already been reported across countries (for example, deficits in executive functions have been reported in: Italy: Nardecchia et al. [26]; the Netherlands: Jahja et al. [27]; the UK: Palermo et al. [15]; the USA: Brumm et al. [25]; Christ, et al. [28]; Diamond et al. [2]; deficits of speed of processing have been reported in Australia: Moyle et al. [10]; Germany: Feldmann et al. [29]; the UK: Channon et al. [30]; Palermo et al. [15]; the USA: Janos et al. [31]). These results suggest that the effects of Phe on cognition are similar across countries, in spite of cultural and environmental differences in the approach to food and feeding. However, there is a lack of studies directly comparing results collected using the same tests and direct comparisons are important to give us confidence that results can be accrued.

The objectives of the present study are twofold: 1. we aim to replicate results previously obtained with a relatively large sample of English AwPKU (*N* = 37) [5,6,15,32] by administering a subset of tasks to a new sample of Italian patients; 2. we aim to demonstrate that the same tasks which are sensitive to blood Phe in English are also sensitive in Italian so that they also demonstrate impairments and relationships with blood Phe levels and Phe fluctuations throughout the lifespan. A comparison between English and Italian PKU samples is particularly relevant given differences in the approach to food and diet in the two countries. Note that there is no issue of validity and specificity in the cognitive assessment of PKU. We do not need to distinguish people with PKU from healthy people. Genetic tests reliably establish the presence/absence of PKU from birth and high-level of Phe are constant in people of PKU if the disease is untreated. What is important, instead, is test-sensitivity to variations in metabolic control so that cognitive outcomes can be properly monitored. This can be demonstrated by showing impairment compared to healthy controls and correlations with metabolic measures.

We assessed metabolic control and cognitive outcomes in 19 Italian early-treated AwPKU and 19 matched controls and compared performance with that of 19 English early-treated AwPKU and relative controls selected from our previously tested cohort [15]. All groups were matched for age and education. Comparisons between PKU groups were in terms of z scores which considered performance in terms of deviations from the relative control groups. In addition, we assessed the sensitivity of our cognitive battery in Italian by assessing correlations with current and historical blood Phe levels. The Italian and English testing batteries were matched as rigorously as possible. In most cases, our tests were exactly the same (the same materials and procedure); those with verbal stimuli were carefully matched for psycholinguistic variables such as word frequency and word length (e.g., Rey AVLT). Similar levels of impairments and similar correlations with Phe levels will demonstrate test reliability and sensitivity for different national PKU samples. It will also give us confidence that, when the same or matched tasks are used, results can be accrued, allowing more power for analyses.

#### **2. Method**

#### *2.1. Participants*

All PKU participants were adults diagnosed soon after birth (2–3 days in Italy and 5–7 days in the UK).

Nineteen Italian AwPKU were recruited from the Clinical Centre for Neurometabolic Diseases in the Department of Human Neuroscience, Child Neurology and Psychiatry Unit at the Sapienza University of Rome. Three participants were currently treated with sapropterin. Nineteen Italian control participants were recruited among students and friends of the researchers. They were matched to the Italian PKU participants for age and education. Among the Italian participants, 4 had a diagnostic blood Phe level > 600 µmol/L but <1200 µmol/L; 15 participants had Phe > 1200 µmol/L at birth.

To allow a direct comparison between an Italian and an English sample, 19 English AwPKU were selected from a larger sample of 37 AwPKU previously tested [15,16,32]. They were all tested at the Inherited Metabolic Disease Unit at the Queen Elizabeth Hospital in Birmingham. They all had Phe > 1200 µmol/L at birth. They were matched one-to-one with the Italian AwPKU for gender, age, and education. Matching was blind to cognitive performance and metabolic control as possible differences were assessed. Thirty English healthy controls were originally recruited through an advertising volunteering website. From this sample, we blindly selected 19 healthy controls matching the English PKU participants for age and education (in number of years).

Power calculations indicated that 20 participants were necessary in the clinical group and 20 participants in the control group for a one tail effect size of 0.8 (consistent with what we found in our previous studies) and =0.05, power (1-error probability) = 0.80. All AwPKU treated in the English and Italian centres were invited to participate and were accepted in the study on a first-come, first-served basis. Recruitment stopped when enough participants were tested. After the Italian PKU participants were contacted, one participant became unavailable and we were left with 19 participants, which still gave our study acceptable power (=0.78).

The English study received NHS ethical approval. The Italian study was approved by the local ethics committee. All participants provided informed consent to the study.

#### *2.2. Ethical Approval*

The study was conducted in accordance with the Declaration of Helsinki. All participants gave their informed consent for inclusion before they participated in the study. In the UK, the protocol was approved by the West Midlands NHS Ethics Committee (rec REC: 10/H1207/115). In Italy the protocol was approved by the Institutional Review Board of "Sapienza"—University of Rome (Project identification code 3629).

#### *2.3. Metabolic Measures*

For both the English and the Italian participants, metabolic measures were taken regularly since birth and extensive records were available. The number of measures did not differ between countries (see Table 1). Blood Phe monitoring was performed on dry blood spot collected after overnight fasting by High Performace Liquid Chromatography until 2010 and then via tandem mass spectroscopy. The laboratories of both centres have adhered and contributed to international quality control systems


**Table 1.** Demographic Variables in Terms of Age, Gender, Years of Education and Metabolic Control across Ages for the Two Matched Groups of Italian and English AwPKU. Blood Phe measured in µmol/L.

AwPKU: adults with PKU, IQ: intelligence quotient, n.s.: not significant, Phe: Phenylalanine.

We averaged metabolic control in three age bands: childhood: 0–10 years old, adolescence: 11–16 years old, and adulthood: 17 years to present. We have also averaged measures throughout the life-time and considered current Phe levels. For the Italian group, current Phe has been measured immediately before the testing session/s or in the preceding few days; for the UK group, current Phe has been measured immediately before the two testing sessions and averaged. We considered two

types of measures: Phe level and Phe fluctuation/variation (we will use the term fluctuation from now). Phe level in each band was calculated by taking the median values for each year and then averaging the year values; Phe fluctuation was calculated by taking the SD for each year and then averaging year values in the band.

#### *2.4. Cognitive Assessment*

Cognitive assessments were carried out in a quiet room at the clinical centres in Birmingham and Rome. The testing session for the Italian participants lasted between 2 and 3 h. The English participants carried out more tests and were tested in two separate sessions of similar length. Testing was carried out by a psychologist or a neuropsychiatrist with neuropsychological training.

IQ was measured, for the Italian participants, using the Wechsler Adult Intelligence Scale-Revised (WAIS-R) [33] and, for the English participants, theWechsler abbreviated scale of intelligence (WASI) [34], which includes the following subtests: Vocabulary, Block Design, Similarities, and Matrix Reasoning. The WAIS-R and the WASI are strongly correlated providing similar validated measures of Full Scale IQ [35]. In addition, participants were given a set of tasks chosen from the larger set of tasks administered in our previous studies [15,16]. We chose tests which either showed a strong difference between PKU participants and controls and/or strong correlations with metabolic measures. We also prioritized tasks with non-linguistic stimuli which did not need adapting across languages. Therefore, we did not include tests of picture naming, reading, spelling, and orthographic knowledge (spoonerisms, phoneme deletions) where speed impairments could be due to a general reduction in speed of processing which was also tapped by visual search. In addition, we did not include tasks where relationship with metabolic measures are modest or absent [5]. Finally, to reduce the number of tasks tapping similar functions, we also did not administer the Tower of Hanoi, lexical learning task, the Stroop, and nonword repetition. Measures of short-term memory (digit span and Corsi span) and a baseline measure of peripheral speed of processing were included for completeness and to confirm or disprove impairments, given mixed results from the literature (for impairments in digit span and nonword repetition see Palermo et al. [15]; for contrasting results see Brumm et al. [25], and Moyle et al. [10]; see also Jahja et al. [27] for deficits with increasing working memory load).

The following cognitive areas were assessed:

#### 2.4.1. Visual Attention

This was assessed with four tasks [15,16]: 1. Simple Detection: Press a response button as soon as a ladybird appears on the screen; 2. Detection with Distractors: Press a button whenever a ladybird appears on the screen alone or with a green bug, in the second part of the task the instruction was changed to press a button whenever a green bug appeared on the screen alone or with a ladybird; 3. Feature Search: Detect a target among distractors not sharing features by pressing a "yes" or "no" button (e.g., a red ladybird among green bugs); 4. Conjunction Search: Detect a target among distractors sharing features (e.g., red ladybird among red bugs and green bugs). Both reaction times (RT from now on) and accuracy measures (error rates) were taken.

#### 2.4.2. Visuo-Motor Coordination

This was assessed with two tasks: 1. Grooved Pegboard Test [36]: Put pegs into the holes of a board using only one hand as quickly as possible (short version with two trials one with the dominant and one with the non-dominant hand for the Italian and English matched samples) and 2. Digit Symbol Task [33]: Fill as many boxes as possible with symbols corresponding to numbers (key with associations remains visible) in 90 s. Trail Making Test A (TMT A) [37,38]: connect circles containing numbers in ascending order of the numbers as quickly as possible.

#### 2.4.3. Complex Executive Functions

This was assessed with four tasks tapping skills such as planning, flexibility, and abstract thinking: 1. The Wisconsin Card Sorting Test (WCST) 64 card version [39]: Discover the rules to match cards from a deck with four reference cards according to the shape, number or colour of the symbols on the card; feedback is provided to allow learning. Flexibility is required when the sorting rule is changed unknown to the participant and the new rule must be discovered. We used three different scores: total errors, number of perseverative responses and number of completed categories. 2. Difference in speed between Trail Making Test B-A (TMT B-A) [37,38]. A involves connecting circles containing numbers in ascending order; B also involves connecting circles in ascending order but alternating between the number and letters contained in the circles. Only time is considered in this test; when occasionally an error is made, it is corrected by the examiner and this affects the time to complete the task. 3 Fluency: For letter fluency: generate as many words as possible starting with a given letter in a minute of time (for Italian: P, F and L; Novelli et al. [40] for English: C, F and L; Benton et al. [41]); for semantic fluency [42,43]: generate as many names of animals as possible in 1 min of time. This requires planning an efficient search through the lexicon.

#### 2.4.4. Short-term Memory/Working Memory

This was assessed with two tasks: 1. Digit Span: Repeat a sequence of digits spoken by the examiner soon after presentation; 2. Corsi Block Tapping Test [44]: The examiner taps a sequence of blocks and the participant must reproduce the sequence in the same order. Span was calculated as the longest sequence which could be repeated correctly (1 point was awarded for each length if all trials are correct; otherwise a corresponding fraction of point was awarded; for digit span where sequences started from length four, sequences of length 1–3 were assumed all correct, unless there were errors with sequences of four digits; in this case, sequences of shorter length were presented).

#### 2.4.5. Sustained Attention

This was assessed with the Rapid Visual Information Processing task (RVP) [45]: detect three target sequences of three digits by pressing the response key when the last number of the sequence appears on the screen. Scores are percentage correct.

#### 2.4.6. Verbal Memory and Learning

This was assessed with The Rey Auditory Verbal Learning Test [46,47] which asks for learning, immediate recall, and delayed recall of a list of 15 words. The list is presented five times and participants are asked to recall the words immediately after each presentation. After the 5th presentation (A5), an interfering list (B1) is presented and participants are asked to recall this list and then, once again, the original list (A6) without a further presentation. Finally, participants are asked to recall the original list after a 20-min filled interval. Our scores include total number of errors across the five learning trials (A1-5); errors in recalling the words after an interfering list (A6); and, again, errors in delayed recall of the original list.

#### 2.4.7. Visual Memory and Learning

This was assessed with the Paired Associates Visual Learning [48]: Learn to associate objects with locations. Z scores for each participant for each task were computed using the relevant control group as a reference point. As well as considering performance in individual tasks, for each participant, we computed two indexes of overall cognitive performance: 1. We averaged z scores in all tasks and 2. We considered the rate of poor scores across all tasks (rate of Z scores => 1.5); this second index is important since an average Z score may mask significant areas of difficulties in a number of tasks (given a profile were some skills are good and others are impaired).

#### **3. Data Analyses**

We used two-tailed *t*-tests to compare the demographics and metabolic measures of the Italian and English AwPKU, as well as to compare the demographics and cognitive performance of each of these clinical groups with the corresponding control group. This allowed us to assess whether the Italian AwPKU demonstrated similar impairments to the English AwPKU [15]. Additionally, we computed Z scores for each PKU participant and each task, using the corresponding control group (subtracting each test value from the corresponding control mean and dividing by the control SD). Average Z scores for Italian and English AwPKU were analyzed using two-tailed *t*-tests to compare size of standardized effects. For this and all other analyses in the paper, we were more interested in comparing patterns across groups than in the significance of an individual measure. However, with the *t*-tests, we have indicated comparisons which remained significant after a Bonferroni correction.

To demonstrate that our battery is as sensitive to impairment in Italian as in English, we carried out Person r correlations between all our cognitive measures and measures of metabolic control, both current and historical. This resulted in a high overall number of correlations for each clinical group (*N* = 144). Individually, each correlation is not very reliable since correlations are notoriously unstable with a small N (19 participants). However, we were not interested in the significance of individual correlations, but, instead, in establishing the sensitive of our tasks to metabolic controls in Italian as in English, across measures. Considering patterns across a large number of cognitive tasks and metabolic measures will reduce error and boost power. We used a χ 2 s to assess the significance of a positive/negative correlations ratio against a chance 50/50 ratio, which is expected if no true relation exists between cognitive performance and metabolic control. In addition, we used a one-sample *t*-test to assess whether the average correlation was significantly different from 0 (see Romani et al., [6] for a similar methodology). We carried out these analyses on both Italian and English PKU samples.

#### **4. Results**

#### *4.1. Demographics*

Demographic and metabolic information for English and Italian PKU groups are shown in Table 1. The Italian and English groups were matched for age, education, and gender. They did not differ significantly for of any of these variables or for IQ. However, the average Phe level was higher in the Italian group, with differences reaching significance for all age bands but adolescence. In addition, Phe levels were more variable in the Italian than in the English group in adulthood and when measured throughout the lifetime.

#### *4.2. Cognitive Performance*

#### 4.2.1. AwPKU vs. Controls

Table 2 shows the performance of the Italian and English PKU groups compared to matched samples of healthy controls. Results show that both Italian and English AwPKU were impaired in a similar range of tasks with good overlap with the English PKU group.

Both Italian and English AwPKU were impaired in IQ measures and showed a reduced speed in allocating visuo-spatial attention and good accuracy in visual search tasks. Both groups showed impairments in tasks tapping visuo-motor coordination (but for the Italian group differences reached significance only for the digit symbol test). The Italian group also showed impairments in the TMT b and a-b and in the fluency tasks and in the RVP task. Both groups showed no difference in the WCST. Finally, both groups performed well in verbal and visuo-spatial short-term memory tasks (digit span, Corsi Block tapping test) and in tasks tapping learning and memory with only a marginal impairment for the Italian AwPKU.


**Table 2.**Cognitive Performance of Italian and English PKU Participants and Healthy Controls Matched for Age and Educations to the Respective Clinical Groups.


**Table 2.** *Cont.*

Italian and English PKU groups are also matched a-priori for age and education. Diff. = Difference. # = comparison which remains significant after Bonferroni correction. WCST: Wisconsin Card Sorting Test, RT: reaction Times, RVP: Rapid Visual Information Processing task, VIQ: verbal IQ, PIQ: performance IQ, FIQ: full IQ.

#### 4.2.2. Italian AwPKU vs. English AwPKU

Table 3 compares standardized performance of Italian and English PKU groups. Despite their worse metabolic control, the Italian group performed significantly worse than the English PKU group in only a few tasks. They were worse on the TMT B and B-A and on letter-fluency, marginally worse on list recall after interference (retention A6), but better on the Corsi Block span. Overall, their performance was numerically worse in terms of average Z score, but this was not statistically significant. When we co-varied concurrent Phe, only the differences in trail making test B-A and Corsi span remained significant (*p* = 0.009 and *p* < 0.001); differences in Trail making B and letter fluency were only marginally significant (*p* = 0.06 and *p* = 0.08).

A detailed comparison of results in visual search tasks highlights the similarity of patterns in the Italian and English PKU groups. Figure 1 shows RTs in the different conditions. Errors were too few for analysis. Across groups, RTs in feature search do not increase with the number of distractors. This is because in this condition the target item "pops out" and a parallel search suffices for a correct answer (generating a flat profile). Instead, across groups, there is a steep increase in RTs with the number of distractors in conjunction search. This is consistent with the need to serially explore the distinct locations of the display in this condition. Moreover, all groups show slower RTs with "No" rather than "Yes" trials, especially in the conjunction search. This is also expected. To decide that an item is not present you need to check all locations in the display; instead to find a target item, only half of the locations need to be checked on average. The PKU groups show the same patterns of the controls, but they are slower in all conditions.

Trials (F (1, 36) = 43.9; *p* < 0.001; η<sup>p</sup> <sup>2</sup> = 0.550). There were also a number of significant interactions: task x distracters (F (2, 72) = 77.7; *p* < 0.001; η<sup>p</sup> <sup>2</sup> = 0.683), because responses were slower with more distracters in the conjunction search, but not in the feature search; distracters x trial (F (2, 72) = 16.0; *p* < 0.001; η<sup>p</sup> <sup>2</sup> = 0.307) because the effect of the number of distracters was more marked in the "no" than the "yes" trials; task x trial (F (1, 36) = 64.8; *p* < 0.001; η<sup>p</sup> <sup>2</sup> = 0.643) because slower responses in the "no" rather than the "yes" trials occurred in the conjunction search, but not in the feature search. Finally, there was a three-way interaction: task x distracters x trial (F (2, 72) = 12.6; *p* < 0.001; η<sup>p</sup> <sup>2</sup> = 0.259) because responses were slower with the number of distracters more in the "no" than the "yes" trials, but only in the conjunction search. Crucially, however, there were no significant interactions within the group.

When PKU participants are compared to controls, differences are constant in all conditions, except in the conjoined task where the English PKU group shows increasing differences from controls with number of distractors: a fanning-out pattern (distracters x group: F (2, 72) = 3.8; *p* = 0.03; η<sup>p</sup> 2 ). Instead, differences from controls always remained stable in the Italian group (F (2, 72) = 0.25; *p*= 0.78). We will comment on this difference in the General Discussion.

Finally, we want to highlight that, as reported for the English sample (Palermo et al., 2017), the variability in cognitive performance in Italian PKU participants is striking. Five PKU participants (26% of the sample) had a normal cognitive profile when compared to the control group (average Z score: −0.2; −0.7−0.1; % z score =>1.5: 5.0; 0–8.3; expected 6.7%; Full IQ = 114; 104–124). These participants had fast speed of processing while maintaining a good accuracy.


**Table 3.** Cognitive Performance of English and Italian PKU Groups in Term of Z Score Computed from Relative Control Groups Matched for Age, Gender, and Education.

Diff. = Difference. # = comparison which remains significant after Bonferroni correction.

**Figure 1.** Performance of the UK and Italian PKU groups and relative control groups on the visual search tasks. Dist: distractors, PKU: Phenylketonuria, RT: reaction Times.

#### *4.3. Cognitive Outcomes in Relation to Metabolic Control*

We ran Bivariate Pearson r correlations between metabolic measures taken at different times during the life span and our adult cognitive measures. To reduce the number of variables per task, we only run correlations for search tasks, with RT measures; for TMT, with the B–A condition; for the WCST, with the total errors; and for the Rey, with total errors over 1–5 trials and with delay recall. Table 4 shows results for the Italian sample. Results for the larger sample of English AwPKU are reported in Romani et al. [5]. We do not report correlations with the English matched sub-sample because correlations with a small sample are notoriously unstable. Thus, results with a smaller sub-sample may differ from the results obtained with a larger sample in potentially misleading ways. However, to compare results for the two languages with samples of an equal size, we also report, with both PKU samples, the % of positive correlations and the average correlation.

The Italian sample had widespread significant correlations with only a few tasks failing to show any significant correlation with any metabolic measure (pegboard task, the semantic fluency task, and the Rey test). Almost all correlation (91%) were positive (132/144; χ <sup>2</sup> = 60.5; *p* <.001). The average correlation across tasks was 0.29 (SD = 0.22), which was significantly different from 0 (one sample *t*-test (142) = 17.0; *p* < 0.001). In the English matched sample (*N* = 19), 69% of correlations were positive (99/144 = 69%; χ <sup>2</sup> = 10.5; *p* < 0.001) and the average correlation was 0.12 (SD = 0.25; one sample *t*-test (142) = 5.7; *p* < 0.001). Taken together, these results indicate that our tests are sensitive to the level of metabolic control in both languages.

Considering the qualitative pattern of correlations in the Italian sample, a number of significant features can be noted which replicate findings from the larger English sample [5]: 1. Adult cognitive performance correlates with current Phe levels, but also with historical Phe records with significant correlations across the life-span; 2. Cognitive performance correlates with metabolic measures in terms of both average Phe levels and Phe fluctuations (average SD per year), this confirms the importance of maintaining not only low average Phe levels, but also of being constant and of avoid Phe peaks; 3. There are interactions between type of function an age when metabolic control is measured: some functions are more effected by childhood Phe measures than by current Phe levels (see visuo-attentional speed). Other functions, instead, show as much of an association with current levels as with historical measures (IQ, visuo-motor control, sustained attention, memory, and learning.


**Table 4.**Correlations Between Phe Measures Across the Life-Span and Cognitive Performance in Different Domains for the Italian PKU Sample (*N*=19).

Highlighted rows compare childhood Phe measures with current Phe. To facilitate interpretation, positive correlations always indicate that high Phe was associated with a worse performance. Thus, for IQ, digit span, Corsi span, and semantic fluency correlations were reversed. Significant measures are in bold. \* = significant < 0.05; \*\* significant < 0.01. Phe SD: Phe fluctuations, FSIQ: full scale IQ, TMT: Trail Making Test, ATTENT: attention, PHE: Phenylalanine.

### **5. Discussion**

There is a lack of studies which have assessed the behavioural effects of metabolic control in individuals with PKU across countries with the same testing materials, comparing sensitivity. This, instead, is important for establishing common PKU testing batteries and to accrue results across centers for a rare disease. In spite of tremendous advances in our knowledge of PKU, we lack a complete and reliable picture of cognitive outcomes in relation to metabolic controls at different ages, which is crucial to establish the efficacy of new therapeutic interventions and to track developmental trends which may demonstrate either cognitive improvements (by bridging developmental gaps) or cognitive deterioration with age (due to abandoning the diet and/or accelerated aging). Our study tested cognitive outcomes in two samples of Italian and English AwPKU closely matched for gender, age, and education, using the same or closely matched materials.

### *5.1. Metabolic Control*

Our results showed better metabolic control in the English than the Italian PKU sample. This could be due to differences in clinical practice (the English group is one of the better controlled groups described in the literature so far, which is likely due to the strong clinical advice received) or to cultural differences in approaches to food across countries. While in Italy, the diet may be less protein-based than in England, the stronger centrality of food in Italian society may make following a separate diet more difficult. Our matched samples are small, and we do not have corroborative information indicating whether following a PKU diet may be easier in England than in Italy. Nonetheless, our results point to variability in metabolic control across groups, suggesting the need to explore possible, modulating socio-cultural factors which could affect clinical management.

### *5.2. Cognitive Impairments*

Our results showed similar patterns of cognitive outcomes and associations with metabolic measures across the Italian and English groups. The Italian PKU participants performed worse than the matched English PKU participants in a few tasks, consistent with their worse metabolic control, but the overall level of impairment and the pattern of spared and impaired functions were similar across groups. The results with an Italian sample confirm patterns of impaired and spared functions previously reported with English participants [15]. They confirm substantial impairments in the following:


The Italian sample showed no impairments in verbal and visual short-term memory. Performance was in fact better than the controls in the Corsi task, confirming the lack of impairment found in Palermo et al., [15]. We have previously found a marginal impairment in digit span. The present results are also consistent with Moyle et al., [49], who found no impairment in the working memory index of the WAIS, and with Brumm et al. [25], who found no impairment in the forward digit span, but an impairment with the backwards span (see also Jahja et al. [27] for deficits with increasing working memory load).

Finally, our results confirm no or marginal impairments in verbal and visual learning (for no impairments in adults see also Channon et al. [52]; Moyle et al. [49]; for a review in children showing mixed results see Janzen and Nguyen, [14]). These results support the idea that AwPKU perform better when they can rely on learning and stored knowledge (see good performance in word spelling; better word than nonword reading speed; and better digit span than nonword repetition; [15,16]).

When we have analysed patterns in visual search tasks, the Italian and English PKU samples also showed similar results. Both groups were slower in conjunction than in feature search tasks and showed increased delays with a number of distractors, especially in conjoined search and target absent conditions. Compared to matched control, both Italian and English PKU samples showed delays across tasks and conditions. However, in conjunction search, the English PKU group showed increased delays with a number of distractors (a fanning out pattern), while the Italian PKU group showed stable differences. In Romani et al. [16], we argued that two different types of difficulties may contribute to a speed reduction in AwPKU: 1. Some specific processing difficulties such as a difficulty in allocating visual attention; this will result in increasing differences from the control with number of distractors; and 2. A general tendency to be more cautious/tentative in returning an answer which will result in a fixed delay across conditions (a flat profile across conditions; a pattern that we have seen with language tasks, see Romani et al., [16]). The Italian PKU sample is relatively small, and it is possible that noise may have obliterated a fanning-out pattern in conjunction search. With this caveat, a stable difference from controls, which is not affected by task difficulty, is more consistent with a delay in making decisions than with processing difficulties.

Taken together, our results highlight how PKU does not impact cognition homogeneously. Some functions are spared while others—like visuo-attentional speed—are severely impaired, showing average Z scores between 1.2 and 1.5 from the control mean (and Z score SDs between 1.2 and 2.3), which indicate that some people with PKU are at the extreme margin of the normal distribution. To consider this variability is important for an accurate evaluation of outcomes in relation to current and future treatment options. Our results also confirm the extreme variability in performance across individual participants. In the Italian sample, 5 participants (26% of sample) showed a completely normal cognitive performance.

Concerning the relationship between cognitive and metabolic measures, our Italian results confirm extensive correlations between current and historical Phe measures and performance in cognitive tasks. Previous studies have also found correlations between current and life-time Phe levels and cognitive functions (see Brumm et al. [25] for correlations with backwards span, fluency, WCST; Smith et al. [50] for a correlation with WCST; Moyle et al. [49] for correlations with TMT and visual memory; Jahja et al. [27] for correlation with visual search tasks with a memory load; see Romani et al. [5], and Hofman et al. [56], for a more detailed review of the literature). Other studies have found differences in cognitive abilities between AwPKU with better versus worse metabolic control. For example, Nardecchia et al. [26] found that patients with a worse metabolic control had a lower IQ and worse performance at the WCST and at the Elithorn's Perceptual Maze Test; Bik-Multanowski et al. [53] found that patients with a worse metabolic control performed worse in RVP, Spatial Span (SSP), Spatial Working Memory (SWM) and Stop Signal Task (SST; see also Brumm et al. [25] and Palermo et al. [15] for results on more comprehensive sets of cognitive tasks).

Our results also confirm that both measures in terms of average Phe and Phe fluctuations correlate with performance [57] and that there are interactions between type of function and age of metabolic control. Some functions are more affected by historical Phe and others by current Phe. Speed in visual attentional tasks is associated mainly with childhood Phe (both average and SD), while tasks tapping visuo-motor coordination (digit-symbol, TMT), sustained attention and memory and learning are also affected by current Phe [5,11].

Our results highlight the importance of using the right cognitive and metabolic measures to assess outcomes. For example, a recent study by Bartus et al. [58] found no correlations between Phe levels (current and life-time) and performance on three tasks of the computerized Cambridge Cognition Test (CANTAB)—Motor screening test, Spatial Working Memory test, and The Stoking of Cambridge—in 47 AwPKU, nor any difference between groups with high vs. low Phe levels. However, we also found no correlations when we used similar tasks with our Italian/English samples (motor speed tapped by simple RT, visual WM tapped by Corsi span) or with larger English sample (the Tower of Hanoi was not tested here; see Romani et al., [5]). This example shows the importance of using both a comprehensive and an ad hoc set of tasks when assessing outcomes and their relationship with metabolic variables. Failing to do so runs the risk of reaching the wrong conclusions regarding the effects of relaxing metabolic control.

#### **6. Conclusions**

It is important to track the cognitive performance of people with PKU across the lifespan. This is the first generation of early-treated AwPKU to move towards middle-age and people with PKU show a tendency to progressively relax the diet with age [55,59–61]. However, we do not know what effects prolonged high-levels of Phe may have on aging brains. Our results suggest that the effect of Phe on the brain is different in childhood to early adulthood [6], but further interactions may be seen later in life. Knowing which functions and relative tasks are most affected and most sensitive to Phe in young adults with PKU provides an important base-line to compare outcomes across the life-span and evaluate the effectiveness of treatment. Our study has contributed to an identification of sensitive tasks by showing consistency across Italian and English samples in the patterns of impaired and spared functions and similar patterns of correlations with metabolic measures and by replicating previous findings. This provides preliminary evidence that common PKU batteries can be used across countries to detect impairments with similar sensitivity. The similarity of the results across the Italian and English PKU samples justified combining results in a single database in a follow-up study, giving us more power to assess the interactions between types of metabolic variable (Phe average vs. Phe fluctuation), age of metabolic control (childhood, adulthood, current) and type of cognitive functions, and more power to assess the relationships between cognitive scores and adherence to metabolic guidelines [6].

#### *Limitations*

The main limitation of our study lies in the small sample of AwPKU tested in Italian. Although this small sample is still sufficient to establish that our tasks are sensitive to metabolic control, larger samples are necessary to compare individual correlations and replicate our preliminary findings that qualitive patterns are the same across languages and nationalities. There was variability in the correlation patterns shown by the Italian and English PKU samples. This is not surprising. The two PKU samples were not matched for metabolic controls. More importantly, correlations are notoriously unstable and have small samples, and this is a major stumbling block for research trying to establish the relationship between metabolic controls and cognitive outcomes in people with PKU. However, it is precisely this limitation that makes it important to establish that tasks are equally sensitive across different nationalities so that results can be accrued. We hope that that our study will be followed by further research to assess the sensitivity of the same tasks across languages in people with PKU.

**Author Contributions:** The authors contributed to the studies as follows: conceptualization, C.R., F.M., F.N., A.M., L.P., V.L; methodology: C.R., F.M., F.N., A.M., L.P., V.L; software: C.R., L.P.; formal analysis: C.R., L.P.; investigation: F.V., N.F., C.C., S.D.L.; resources: V.L, C.R.; writing—original draft preparation: C.R.; writing—review and editing: C.R., F.M., F.N., A.M., L.P., V.L.; supervision: C.R., F.M., F.N., A.M., L.P., V.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We would like to thank Tarekegn Gerberhiwot for having allowed access to the PKU participants, Cecilia Guariglia for having facilitated tested of the Italian control participants and Andrew Olson for helpful comments on a version of the manuscript. The original cohort of English patients was supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Programme granted to Liana Palermo under the supervision of Cristina Romani and by a grant of the University Hospital Birmingham Charity to Tarekegn Gerberhiwot.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **The Potential Roles of Blood–Brain Barrier and Blood–Cerebrospinal Fluid Barrier in Maintaining Brain Manganese Homeostasis**

**Shannon Morgan McCabe and Ningning Zhao \***

Department of Nutritional Sciences, The University of Arizona, Tucson, AZ 85721, USA; morgans3@email.arizona.edu

**\*** Correspondence: zhaonn@email.arizona.edu

**Abstract:** Manganese (Mn) is a trace nutrient necessary for life but becomes neurotoxic at high concentrations in the brain. The brain is a "privileged" organ that is separated from systemic blood circulation mainly by two barriers. Endothelial cells within the brain form tight junctions and act as the blood–brain barrier (BBB), which physically separates circulating blood from the brain parenchyma. Between the blood and the cerebrospinal fluid (CSF) is the choroid plexus (CP), which is a tissue that acts as the blood–CSF barrier (BCB). Pharmaceuticals, proteins, and metals in the systemic circulation are unable to reach the brain and spinal cord unless transported through either of the two brain barriers. The BBB and the BCB consist of tightly connected cells that fulfill the critical role of neuroprotection and control the exchange of materials between the brain environment and blood circulation. Many recent publications provide insights into Mn transport in vivo or in cell models. In this review, we will focus on the current research regarding Mn metabolism in the brain and discuss the potential roles of the BBB and BCB in maintaining brain Mn homeostasis.

**Keywords:** manganese; blood–brain barrier; blood–cerebrospinal fluid barrier; choroid plexus

#### **1. Manganese Dyshomeostasis and Neuropathological Consequences**

Manganese (Mn) is essential for life as it is necessary for the normal function of several enzymes, including the antioxidant enzyme Mn superoxide dismutase (MnSOD) [1] and the neurotransmitter synthesis enzyme glutamine synthetase [2]. Since adequate Mn is easily obtained through a healthy diet, Mn deficiency is uncommon. However, Mn overload occurs more frequently and becomes a public health concern. Exposure to high levels of Mn in occupational environments such as mining, welding, and dry cell battery production can lead to manganism, which is a disorder characterized by serious and irreversible neurological symptoms similar to those seen in Parkinson's disease. Early symptoms of manganism caused by occupational hazards include neurobehavioral changes such as impulsiveness and irritability, followed by changes in gait and difficulty with speech as the disease progresses [3]. High Mn levels in local drinking water, along with elevated Mn in blood and hair samples, reveals a correlation between higher Mn levels and decreased memory, verbal, and overall IQ scores [4]. Elevated environmental Mn exposure in children is also correlated with poorer academic achievement [5], altered performance on visual perception and memory tasks [6], and reduced Full Scale IQ [7].

In patients with mutations in Mn-transport proteins, excess Mn accumulates in the blood and brain, causing neurological symptoms. Blood Mn levels in healthy individuals is <320 nmol/L, while patients experiencing neurological symptoms of Mn overload have levels exceeding 2500 nmol/L [8,9]. Additional data from patients with inherited disorders of Mn homeostasis have been recently summarized [10]. Individuals can also receive excess Mn from environmental sources. A group of people living in an area with high Mn in drinking water (1.8–2.3 µg/mL) experienced many of the neurological symptoms related

**Citation:** McCabe, S.M.; Zhao, N. The Potential Roles of Blood–Brain Barrier and Blood–Cerebrospinal Fluid Barrier in Maintaining Brain Manganese Homeostasis. *Nutrients* **2021**, *13*, 1833. https://doi.org/ 10.3390/nu13061833

Academic Editor: M. Hasan Mohajeri

Received: 15 April 2021 Accepted: 25 May 2021 Published: 27 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

to manganism, such as tremors, gait disturbances, and memory dysfunction [11], thus highlighting the dangers of excess Mn to neurological health.

μ

Older adults are also at risk of the neurological effects of excess Mn in the brain. Alzheimer's disease (AD) and related dementias are a group of neurological disorders that first present as cognitive impairment in aging individuals. There is no known cause for late-onset AD, but environmental pollutants such as heavy metals are thought to be a contributor [12]. In the brain, reactive oxygen and nitrogen species (ROS/NOS) are normally produced at manageable levels during oxidative phosphorylation. MnSOD is an antioxidant enzyme that requires Mn, but excess Mn reduces its antioxidant activity. In the brain of a patient with AD, there is a decrease of MnSOD activity and increased oxidative stress [13,14]. In non-human primates, chronic Mn exposure induced amyloidbeta precursor-like protein 1 expression and increased the formation of amyloid plaques, which is one of the main neuropathological hallmarks of AD [15,16]. A recent study used a transgenic mouse model of AD and exposed subjects to additional Mn via drinking water (0.36 mg/mL) over five months [17]. At the end of the study, mice consuming Mn-treated water had more beta amyloid deposition in the cortex and hippocampus than untreated transgenic mice. This result shows that Mn consumption may contribute to the severity of AD. In another study, mice were administered daily MnCl<sup>2</sup> doses of either 15 mg/kg or 60 mg/kg intraperitoneally. The study concludes that increased Mn exposure is correlated with increased amyloid-beta in the blood and decreased cognitive test scores in mice [18]. These results suggest that brain Mn dyshomeostasis may be a factor in the development of AD.

#### **2. Structure of the Brain Barriers**

The brain has developed physiological barriers to selectively restrict the exchange of ions and solutes between the blood and brain, allowing a tight regulation of the brain microenvironment for proper neuronal function. In order to enter the brain microenvironment, Mn from the systemic circulation has to cross either of the two strictly controlled blood–brain interfaces: the blood–brain barrier (BBB) and the blood–cerebrospinal fluid (CSF) barrier (BCB). Therefore, the BBB and BCB are the points of restriction for Mn entering the brain from systemic circulation (Figure 1). The accumulation of Mn within the brain and the export of excess Mn back into blood circulation occurs mainly across these two barriers.

**Figure 1.** Localizations of the brain barrier interfaces. The blood–brain barrier (BBB) is localized to the microvasculature of the central nervous system and separates the lumen of cerebral blood vessels and brain parenchyma. Neurons and glia are found in the CNS parenchyma and thus protected from the periphery by the BBB. The blood–CSF barrier (BCB) is formed mainly by the choroid plexus epithelium located between choroid plexus capillaries and the CSF. Materials transported through the choroid plexus epithelium reach the CSF, where they can diffuse into the brain parenchyma.

The unique structures of the brain barriers provide insights into which cell types might express metal transporters. Further, cell models of the BBB and BCB may reflect the physiological structure and features of each barrier. Brain vasculature delivers oxygen and nutrients throughout the brain and shuttles toxins and unneeded materials away from the central nervous system. Unlike other organs, the exchange of molecules between the blood vessel and the brain environment is tightly regulated to prevent the infiltration of harmful pathogens, toxins, and immune factors. The BCB also restricts the movement of molecules between the blood and CSF. In essence, the brain environment beyond the blood vessel or CP epithelium is separated from general blood circulation.

#### *2.1. Structure of the BBB*

Regulatory control of the BBB is provided by specialized barrier cells and their unique structures and junction proteins. The BBB primarily consists of three unique cell types (Figure 2A): endothelial cells of the brain blood vessels, astrocytic end feet encasing the endothelium, and pericytes that form a basement membrane between the blood vessel and astrocyte [19,20]. Endothelial cells of the BBB are polarized, with the abluminal surface toward the brain environment, and the luminal surface facing the blood vessel lumen (Figure 2A). These endothelial cells are linked by tight junctions and adherens junctions to prevent the paracellular movement of water-soluble molecules.

**Figure 2.** Cellular structures of the brain barriers. (**A**) The BBB is composed of endothelial cells (endo) of the brain blood vessel, and it is supported by pericytes, basement membrane proteins (green dashed line), and astrocytic end feet. The luminal (LM) side of BBB endothelial cells faces the inside of the blood vessel. It is also referred to as the apical side. The abluminal (AB) side faces the brain parenchyma and can exchange between the endothelial cell and the astrocytic end foot or brain extracellular space. It can be considered the basolateral side of BBB endothelium. (**B**) The BCB is made up of choroid plexus epithelial (CPE) cells connected to each other by tight junctions and attached to the blood vessel via basement membrane proteins (green dashed line). The apical (AP), or CSF-facing side of the CPE expresses transporters necessary for the secretion of CSF. On the basolateral (BL), or blood-facing side, CPE cells exchange materials with circulating blood, since endothelial cells in the CP lack tight junctions and permit larger molecules to diffuse.

Tight junction proteins exist almost entirely on the interior, protoplasmic face of the endothelial cell membrane. One group of these proteins are zonula occludens-1 and -2 (ZO-1, ZO-2). Both ZO proteins are required for the formation of tight junction strands between endothelial cells in the BBB [21,22]. Junctional adhesion molecule (JAM) proteins are another group of tight junction proteins [23]. Of the three JAM proteins found in the BBB, no individuals appear to be necessary for BBB integrity [23]. However, JAM proteins are highly enriched in BBB tight junctions and are responsible for the apical–basal polarity of endothelial cells, and therefore contribute to BBB formation [24]. Other components of BBB tight junctions are the claudins and occludin. Claudin-5 is the only protein of its family that appears to be localized to the BBB and contributes to barrier integrity [23]. It is also particularly enriched in the brain endothelium over other peripheral blood vessels, indicating the importance of claudin-5 in the formation of the BBB. Similar to claudin, occludin is an endothelial transmembrane protein. Mice with occludin deficiency had

increased calcium precipitation in the brain despite normal serum calcium concentration, suggesting that occludin may be necessary for BBB tight junction integrity [25].

Adherens junctions between endothelial cells of the BBB help maintain the integrity of this barrier by regulating the adhesion between cells and controlling the flow of molecules between the blood and brain [26]. Adherens junctions are primarily composed of vascular endothelial cadherins, catenins, and nectins. Cadherins interact with catenins to facilitate their linkage to the actin cytoskeleton, forming the cadherin-based adhesions between the BBB endothelial cells [23], while nectins promote the establishment of endothelial apical–basal polarity and contribute to adherens junction integrity [27].

Each of these junctions exist between endothelial cells of the BBB, but other cell types are necessary for the barrier structure and function. Pericytes are found in the basement membrane of capillaries and surround the vessel. Differences in pericyte population sizes suggest that pericytes are directly involved in BBB permeability but do not alter tight junction formation [28]. Contractile smooth muscle cells fully surround arterioles to provide blood flow control [29]. Both pericytes and smooth muscle cells assist in the structural development of the brain blood vessels [23]. Surrounding the majority of the abluminal blood vessel are astrocytic end feet. Astrocytes associated with BBB endothelial cells increase the integrity of the BBB by decreasing the permeability of tight junctions [19,30]. Astrocytes function as an extensive network and interact with each other via gap junctions to coordinate ion changes [31], while their end feet are specifically responsible for the exchange of ions and molecules with the blood vessel in order to maintain ion homeostasis [32]. A basement membrane layer fills the gap between the endothelial cells and astrocytes, with pericytes and smooth muscle cells embedded within. Pericytes and endothelial cells form a 3D structure of laminins, nidogens, collagens, and heparan sulfate proteoglycans [33,34]. Communication and transport between the blood and astrocytes occurs through this matrix.

In most other tissues, blood vessels have small gaps, or fenestrations, between endothelial cells to allow larger molecules to cross from blood to tissue. The BBB is considered a physical barrier due to its lack of fenestrations [20], presence of tight junctions, and lack of permeability to large molecules [35]. Transport of hydrophilic molecules, such as glucose and metal ions, requires specific transporters to cross the endothelial membrane, while large molecules can cross via receptor-mediated endocytosis [20].

This highly selective barrier exists in all brain blood vessels, with the exception of the vessels in the meninges and those near the circumventricular organs. The pituitary and pineal glands, as well as the median eminence, paraphysis, and area postrema, possess a less restrictive barrier in order to allow signaling molecules and hormones to reach specific brain areas, without crossing the BBB into off-target areas [36].

#### *2.2. Structure of the BCB*

The other major brain barrier is the BCB, which is localized to the choroid plexus (CP) within the four brain ventricles (Figure 2B). CP tissues in the left and right lateral, and third ventricles are made up of epithelial cells surrounding the anterior choroidal and posterior choroidal arteries, while the fourth ventricle epithelium receives blood flow from the anterior and posterior inferior cerebellar arteries [37]. A thin endothelial basement membrane lies on the abluminal side of the blood vessel [38]. In contrast to the blood vessels forming the BBB, the capillaries of the CP are highly fenestrated and lack tight junctions to connect the endothelial cells, allowing the movement of larger molecules from the blood vessel to the CP tissue. These molecules first reach the stroma, which is a layer of fibroblastic mesenchymal-like cells that surround the CP blood vessels [37,39,40]. Leukocytes, macrophage, and dendritic cells are known to migrate to this cell layer from the blood vessel before being transported across the epithelial layer into the brain [40–42].

The outermost layer consists of polarized CP epithelial cells that are connected to the basement membrane and stromal layer on their basolateral side, allowing the cells to interact with systemic blood circulation while the apical, CSF-facing side is responsible for producing CSF and exchanging materials with the ventricles [40,43]. The presence of microvilli on the apical brush border of the epithelial cells increases the surface area and facilitates the transport of molecules into the ventricle [44].

CP epithelial cells are connected by tight junctions, thus restricting free passage of large or hydrophilic molecules into and out of the brain. While knowledge of CP tight junction proteins is less complete than that of the BBB, it is currently understood that many tight junction proteins of the BBB are also expressed in CP epithelial cells. The decreased occludin level induces epithelial permeability to larger molecules, suggesting that occludin may be a necessary component for the formation of CP tight junctions to block the transfer of large molecules across this barrier [45]. The expression of various claudins in the CP epithelium may be specific to developmental stages and species [46,47]; however, consistent reports of human and murine tight junctions show that claudin-1, -2, and -3 can be detected in CP epithelial tight junctions [48,49]. As in the BBB, the epithelial cells of the BCB express intracellular accessory protein ZO-1 [50] that seems to be required for tight junction integrity, since decreases in ZO-1 expression by inflammation cause an increase in the BCB permeability [51]. Lining all other surfaces of the ventricular walls are ependymal cells—cuboidal epithelial cells that lack tight junctions and are permeable to macromolecules [52]. Since the only structure between the CSF and brain parenchyma is this permeable ependymal cell barrier, molecules in the CSF could enter brain parenchyma by diffusion. Thus, the BCB function is carried out mainly by the single layer of CP epithelial cells and the tight junctions that link them [53,54].

#### **3. Mn Homeostasis at the Brain Barriers: Evidence of Involved Metal Transporters**

*3.1. Potential Roles of Iron Transport Pathway Proteins in Mediating Mn Delivery at the Brain Barriers* 3.1.1. Transferrin (Tf)

Many divalent metals share the same set of transporters. Studies of iron transport and absorption led to the first understandings of Mn homeostasis, particularly in studies of transferrin/transferrin receptor 1 (Tf/TfR1) and divalent metal transporter-1 (DMT1). The Tf cycle is the primary pathway for cells to take up iron. In this pathway, the circulating Tf carries Fe3+, followed by binding to the cell surface TfR1 and subsequent invagination into intracellular vesicles. The acidic pH inside the vesicles causes the release of Fe3+ from Tf. Then, Fe3+ is reduced to Fe2+ and transported into the cytoplasm via DMT1 [55,56]. In addition to Fe3+, Tf can bind to trivalent Mn (Mn3+) [57,58] because Mn3+ is very similar in structure to Fe3+ [59]. Moreover, Mn and Fe accumulate in many of the same brain areas during overload conditions [60]. Therefore, the transport of Mn is presumably tied to the proteins involved in iron transport, including Tf. However, it has been shown that mice with Tf deficiency had similar levels of Mn in the brain compared to the wild-type animals, indicating that Tf is not necessary for the delivery of Mn to the brain [58].

#### 3.1.2. DMT1

As mentioned above, DMT1 is a metal transporter that mediates the efflux of divalent metals from a vesicle to the cytoplasm. DMT1 functions optimally at pH 5.5, but its functionality in cells at pH 7.4 has also been observed [61]. DMT1 can localize to the plasma membranes of enterocytes or hepatocytes in high- or low-Fe conditions, respectively; whereas in regular dietary conditions, DMT1 remained primarily in the cytoplasm [62]. While DMT1 can adapt to changing substrate availability and subcellularly localize accordingly, its expression in the brain barriers is low. DMT1 mRNA expression is very low in isolated rat brain capillaries [63] and brain endothelial cells in culture [55]. In addition, protein expression is not detectable in brain endothelial cells of adult or early postnatal mice [64]. In a developmental study carried out in rats, the expression of DMT1 was detected by immunohistochemistry in the choroid plexus during early postnatal days, with

increased expression at postnatal day 15. DMT1 was detected when staining cerebral blood vessels, but it aligned very closely with astrocyte localization; therefore, the transporter could be present on either the endothelial or glial cells [65]. In contrast, adult rats appear to express DMT1 protein in the CP epithelial cells, but not in microvascular endothelial cells [66], and the expression of DMT1 in CP epithelial cells was observed primarily in the cytoplasm.

A study of brain microvascular endothelial cells of human origin (hBMVEC) provides an analysis of time-dependent uptake of <sup>54</sup>Mn2+ that increased in the presence of a clathrin-dependent endocytosis inhibitor [67], demonstrating that the receptor-mediated endocytosis in the Tf/TfR1 and DMT1 pathway is not involved in Mn transport in this BBB cell model. Moreover, in Belgrade rats that lack functional DMT1, brain Mn levels remained normal in the olfactory bulb, cortex, striatum, hippocampus, and cerebellum, while Fe levels decreased in all brain areas tested [68], suggesting that DMT1 is not required for Mn delivery into the brain.

#### *3.2. ZIP- and ZnT-Family Transporters*

#### 3.2.1. ZIP8 and ZIP14

ZIP14 and ZIP8 are two recently identified members of the Zrt- and Irt-like protein family of metal transporters. Both proteins have been investigated for their roles in brain Mn homeostasis [10,69–72].

In polarized HIBCPP cells, a cell model for the BCB epithelium, ZIP14, was enriched on the basolateral membrane, while ZIP8 was enriched on the apical membrane [73]. The knockdown of ZIP14 or ZIP8 using siRNA-mediated technology led to a decrease in <sup>54</sup>Mn accumulation in HIBCPP cells, although the decrease in <sup>54</sup>Mn accumulation was much greater with ZIP14 knockdown [73]. These results suggest that both ZIP14 and ZIP8 are involved in Mn uptake in this CP epithelial cell model, which is consistent with previous studies on epithelial cell models of intestine [70], lung [74], and liver [75].

In human primary brain microvessel endothelial cells (hBMVEC), a cell model for the BBB endothelium, the expression of both ZIP14 and ZIP8 were identified [67]. The uptake of <sup>54</sup>Mn was dependent on both ZIP8 or ZIP14, with significantly decreased <sup>54</sup>Mn accumulation when one or both proteins were knocked down. In contrast to the expression pattern observed in HIBCPP cells, ZIP14 and ZIP8 were localized to both sides of the polarized hBMVEC cells, where both proteins seem to be involved in apical-to-basolateral and basolateral-to-apical transport of Mn. Flux in the basolateral-to-apical direction was more prominent, modeling the movement of Mn from the brain to the blood through the BBB.

These studies using cell models of the BCB and BBB may provide insights into how Mn is transported within these two barriers. By identifying the polarized localization of ZIP14 and ZIP8 in CP-derived cells, we can begin to understand how Mn is transported at the BCB. Apical ZIP8 expression in HIBCPP cells suggests that Mn could be transported from the CSF into the epithelial cells to facilitate apical-to-basolateral movement of Mn out of the brain. In the same way, basolateral ZIP14 expression in CP-derived epithelial cells indicates that ZIP14 could be involved in the blood-to-brain movement of Mn via import of Mn from the blood to the epithelial cell of the BCB. The uptake experiments in hBMVEC endothelial cells suggest that both ZIP8 and ZIP14 play a significant role in Mn uptake into endothelial cells of the BBB. The basolateral-to-apical flux of Mn in these cells translates to a brain-to-blood movement of Mn in vivo. Thus, the BBB could have a considerable role in Mn clearance from the brain, dependent on coordinated transport by ZIP8 and ZIP14. A future study in the CP-derived HIBCPP cells could be useful to indicate the direction of Mn transport in cells with polarized expression of Mn transporters.

#### 3.2.2. ZnT10

ZnT10 is a member of the Zinc Transporter family proteins. Patients with an inherited homozygous *ZnT10* mutation resulting in a non-functional ZnT10 protein exhibit high Mn levels in the blood and brain, as well as Mn toxicity-induced dystonia [76–78]. In cell culture studies, ZnT10 appears to be a Mn efflux transporter [79] expressed on the surfaces of enterocytes [78] and neuronal cells [80]. In both humans and mice, ZnT10 was highly expressed in the brain, liver, and intestine [78,80,81].

There is evidence showing that ZnT10 mRNA is expressed in the CP in rats [82], but there is no evidence to show that ZnT10 is expressed in the brain microvessels at either the gene or protein level. A recent study used pan-neuronal/glial *Znt10* knockout mice and detected no difference in brain Mn levels with standard dietary conditions [78]. In this study, *Znt10* knockout mice lack the protein in the vast majority of brain cells, including all neurons, astrocytes, and oligodendrocytes. As an efflux transporter, ZnT10 would likely protect neurons and glia from high Mn levels when overloaded, but a lack of increased Mn in the knockout mice indicates that these brain cells are not accumulating Mn, at least in normal conditions. Thus, a normal amount of Mn was circulating in the interstitial fluid and CSF, regardless of ZnT10 expression. This finding provides key information about Mn balance within the brain parenchyma, but the brain environment is controlled by the BCB and BBB. Therefore, Mn levels within the brain may not change without a change in transporter expression in either of the brain barriers. Interestingly, *Znt10* neuronal/glial knockout mice exposed to high dietary Mn experienced a greater increase of Mn in specific brain areas compared to exposed wild-type littermates. This result suggests that there is less Mn transported out of the brains lacking ZnT10, which could indicate that the efflux transporter is normally localized to either the brain endothelium or CP epithelium.

The expression pattern of ZnT10 in the BBB or BCB is unknown. As an efflux transporter, cell type localization of ZnT10 is necessary to understand which barrier is responsible for Mn efflux. Additionally, ZnT10 is likely polarized to either the basolateral or apical surfaces of the epithelial or endothelial cells. If polarized, its location would indicate whether ZnT10 is responsible for brain Mn accumulation or clearance. In addition, cell culture models of BBB and BCB are ideal to elucidate Mn transport mechanisms at the cellular level. Uptake and transport studies, such as those completed in previous studies of ZIP8 and ZIP14, would indicate if ZnT10 is necessary for normal Mn accumulation or efflux. Additionally, transport studies in polarized epithelial or endothelial monolayers would confirm the direction of Mn transport to which ZnT10 contributes.

#### *3.3. ATP13A2*

Another transporter associated with brain Mn homeostasis is ATP13A2. Mice with *Atp13a2* knockout accumulated more Mn in the brain compared to wild-type mice after intraperitoneal administration of MnCl<sup>2</sup> [83]. Since brain Mn levels tend to rise when the blood levels of Mn increase, a future investigation should report blood or serum levels in order to further understand the location of accumulated Mn in *Atp13a2* knockout mice. While knockout of *Atp13a2* causes brain Mn accumulation, overexpression of ATP13A2 in HeLa cells and nematode dopamine neurons had a protective effect against high Mn exposure [84]. Taken together, ATP13A2 appears to have a role in Mn homeostasis within the brain, but it is unclear how it could act as a Mn transporter at the BBB or BCB. To date, there are no publications showing evidence of ATP13A2 in human or mouse brain endothelium or choroid plexus tissue. Future studies of ATP13A2 should identify the transporter's tissue and membrane localization within the brain barriers to determine if the protective effect of this transporter against high Mn exposure is applicable to CP epithelial cells and brain endothelial cells. Major evidence of the metal transporters' involvement in Mn homeostasis at the BBB and BCB is summarized in Table 1.


**Table 1.** Evidence of metal transporters involved in Mn homeostasis at the brain barriers.

#### **4. Brain Mn Accumulation Is Likely to Occur via the BCB**

The BBB and BCB are required to maintain the normal physiological conditions of the central nervous system. These two barriers have distinct but overlapping roles in the exchange of material from the blood to brain, as demonstrated by the difference in transporter expression and transport activity in each barrier. For example, both the CP and the brain endothelium express glucose transporters that deliver energy to the brain via facilitated diffusion [85]. Glucose is transported into the brain through both barriers, although it is estimated that the BCB imports only about 1/100th of the glucose that the BBB transports [86,87]. This pattern of uneven transport may also be applicable to Mn distribution into the brain.

A cell culture study of Mn transport across porcine BCB and BBB models indicated that the BCB is likely the primary route for brain Mn uptake. First, uptake studies indicated that CP epithelial cells accumulate nearly three times more Mn than endothelial cells when exposed to the same amount of MnCl<sup>2</sup> in media. Second, in a Transwell model of polarized cells with MnCl<sup>2</sup> added to each side, it was found that epithelial cells accumulated significantly more Mn in the apical chamber, suggesting that CP epithelial cells predominantly transport Mn in the basolateral-to-apical direction. In the BBB model, endothelial cells did not accumulate more Mn in one chamber than the other, indicating that the BBB transports Mn in both directions equally [88]. A future study might compare these findings with lower Mn concentrations that could be more relevant to physiological or Mn-overload conditions. Nevertheless, the results from this study provide valuable information about the activity of Mn transport across the brain barriers. Importantly, if these results reflect the in vivo behavior of the BCB, increased blood Mn would cause higher basolateral-to-apical Mn transport through the CP epithelium.

In vivo studies also provide evidence for the primary role of the BCB in brain Mn uptake. Brain Mn-mapping studies carried out in animals using peripheral Mn2+ administration followed by enhanced magnetic resonance imaging suggest that the entry of Mn into the CNS occurs predominantly through BCB. First, in mice, 2 h after intraperitoneal MnCl<sup>2</sup> injection, the Mn signal was first enhanced in the CSF-containing ventricles. This ventricular signal cleared over the next 24 h accompanied by a gradual increase of parenchymal Mn intensity. A close examination of different brain regions revealed that the Mn signal was highest in areas immediately adjacent to the CSF-containing ventricles, while the signal intensity steadily decreased with increasing distance from the ventricles [89]. Second, in rats, within 5 min of MnCl<sup>2</sup> injection through the tail vein, the Mn signal was first enhanced in the choroid plexus. At 10 min, the signal diffused to the entire CSF-containing ventricles, and by 100 min post injection, the Mn signal spread into the periventricular tissues that are in contact with the CSF [90]. Third, in marmosets, 1.5 h after the start of MnCl<sup>2</sup> infusion through the tail vein, the Mn signal was initially enhanced in the CP, and at the 2.5 h time point, the signal was detected in the parenchyma surrounding the ventricles. In contrast, throughout the entire 6 h infusion course, no Mn signal was detected in brain regions that are not adjacent to the ventricles [91].

These findings in cell models and animals suggest that a main route for Mn uptake into the brain is from the CP, through the CSF, and then to the brain parenchyma. The cell culture studies suggest that the BBB has a role in Mn transport, but it does not cause the accumulation of Mn in the brain from the blood. Meanwhile, the BCB preferentially transports Mn from the blood into the brain, potentially contributing to brain Mn overload. Since the brain endothelial cells appear to transport Mn in both directions, and CP epithelial cells transport more Mn into the brain than into the blood, future studies in cell models or in vivo could investigate transporters involved in unidirectional or bidirectional transport of manganese across endothelial or epithelial cells.

#### **5. Future Directions**

To further understand Mn homeostasis across the brain barriers, more in vitro models of the BBB and BCB that reflect the barrier qualities of each cell layer need to be developed. Cells modeling the BBB or BCB must polarize, form tight junctions, and prevent the diffusion of large or hydrophilic molecules. Such models are necessary to identify the efflux and influx of Mn through the cells of the brain barriers at different Mn concentrations. For example, Mn accumulation studies in animals indicated that Mn crosses the choroid plexus and quickly travels into the CSF, suggesting a major role for the BCB in Mn uptake. Cellular transport experiments in CP epithelial cells could distinguish how Mn is transported across the basolateral and apical membranes and would bolster the conclusion that the BCB is

the primary site of Mn absorption into the brain. Due to the growing knowledge of Mn metabolism, modulators of Mn transport may be developed in the near future, but no such technology exists at this time. Fundamental research of brain Mn homeostasis may eventually facilitate the development of methods to control the balance of metals in the brain to limit the negative effects of excess Mn.

In vivo research is a necessary step to establish Mn transport mechanisms, but there are a few limitations of animal research in this area. To study Mn transport, there is the difficulty of identifying where Mn is concentrated within the brain parenchyma. In Mn overload conditions, it is unknown whether Mn accumulates in neurons and glia of specific brain regions or within the interstitial fluid and CSF. Most publications report brain Mn levels as the level of Mn in the whole brain homogenate, making it difficult to distinguish between Mn accumulating in brain cells or CSF and interstitial fluid. Future studies are needed to understand this important distinction. To function and fire quickly, neurons rely on steep ionic gradients between their intracellular environment and the surrounding interstitial fluid. Since the concentrations of Ca2+, Na<sup>+</sup> , K<sup>+</sup> , and Cl<sup>−</sup> would be vastly different when sampling either neurons or the extracellular fluid, we could logically understand that levels of other charged ions such as Mn2+ would be different within the neuron or out in the interstitial fluid. Additionally, astrocytes are known to release and take up ions and nutrients, while the brain has changing demands for these materials. Astrocytes may sequester Mn intracellularly or release it back into the interstitial fluid, leaving the total brain Mn concentration unchanged. When using a whole brain homogenate to measure metal levels, the extracellular environment around the BCB cannot be sampled separately. To accurately reflect the Mn concentrations on each side of the BCB, Mn levels in both CSF and blood can be measured. In addition, knowledge of Mn accumulation in separate brain compartments, as well as improved understanding of transporter expression in human and animal tissues, will help make significant advances in the field of Mn homeostasis within the brain barriers.

**Author Contributions:** Conceptualization, S.M.M., N.Z.; writing—original draft preparation, S.M.M.; writing—review and editing, S.M.M., N.Z.; supervision, N.Z.; project administration, N.Z.; funding acquisition, N.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** Research reported in this publication was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) and the Office of Dietary Supplements (ODS) of the National Institutes of Health (NIH) (R01DK123113). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Review* **The Protective and Long-Lasting E**ff**ects of Human Milk Oligosaccharides on Cognition in Mammals**

#### **Sylvia Docq 1 , Marcia Spoelder 1 , Wendan Wang <sup>2</sup> and Judith R. Homberg 1, \***


Received: 30 September 2020; Accepted: 19 November 2020; Published: 21 November 2020

**Abstract:** Over the last few years, research indicated that Human Milk Oligosaccharides (HMOs) may serve to enhance cognition during development. HMOs hereby provide an exciting avenue in the understanding of the molecular mechanisms that contribute to cognitive development. Therefore, this review aims to summarize the reported observations regarding the effects of HMOs on memory and cognition in rats, mice and piglets. Our main findings illustrate that the administration of fucosylated (single or combined with Lacto-N-neoTetraose (LNnT) and other oligosaccharides) and sialylated HMOs results in marked improvements in spatial memory and an accelerated learning rate in operant tasks. Such beneficial effects of HMOs on cognition already become apparent during infancy, especially when the behavioural tasks are cognitively more demanding. When animals age, its effects become increasingly more apparent in simpler tasks as well. Furthermore, the combination of HMOs with other oligosaccharides yields different effects on memory performance as opposed to single HMO administration. In addition, an enhanced hippocampal long-term potentiation (LTP) response both at a young and at a mature age are reported as well. These results point towards the possibility that HMOs administered either in singular or combination forms have long-lasting, beneficial effects on memory and cognition in mammals.

**Keywords:** human milk oligosaccharides; cognition; brain development; animal behaviour; fucosyllactose; sialyllactose; long term potentiation

#### **1. Introduction**

The natural composition of breast milk is well recognized as the golden standard of infant nutrition [1] and is associated with long-term health benefits [2–10]. Studies have shown that exclusive breastfeeding is accompanied by a reduced risk for developing medical conditions during childhood such as gastrointestinal infections (e.g., necrotizing enterocolitis) [5,6]. Indications that breastfeeding confers protective effects in the onset and course of allergic diseases such as atopic dermatitis, food allergy and asthma have also emerged over the recent years [7–9]. Such protective effects of breastfeeding have been attributed to multiple factors related to the gut, as it is found that breastfeeding can improve immune functioning, promoting a healthy gut microflora [11]. Apart from the gut, bioactive components within breast milk such as the adipokines (e.g., leptin, ghrelin) help regulate appetite control and energy intake. Breast milk also contains growth factors, such as neuronal growth factors (NGF) and epidermal growth factors (EGF), which exert trophic effects on the neonatal nervous system and enhance gastrointestinal mucosal maturation respectively [11–13]. In recent years, the mental health benefits that breastfeeding provides have garnered much more attention in

neuroscientific research. Notably, breastfeeding is associated with improved cognitive development, as demonstrated by improved IQ scores [14] and a reduced risk of childhood behavioural disorders [15,16]. These findings also coincide with studies showing enhanced brain development parameters, such as white matter development in frontal and temporal regions [17] and maturation of the basal ganglia and thalamus [18]. On the whole, these studies indicate that there are clear developmental and cognitive benefits related to breastfeeding and breast milk, which raises the question: which breast milk factors facilitate cognitive development?

Breast milk is a complex liquid which contains many different lipids (such as the Milk Fat Globule rich in phospholipids and long chain fatty acids), an assortment of vitamins (Vitamin A, B, C, D K), sialic acid (both in free form and bound to oligosaccharides, glycoproteins and glycolipids) and other biologically active components, some of which affect neurodevelopment [19–22]. Of particular interest to infant nutrition and development are the Human Milk Oligosaccharides (HMOs). These non-digestible carbohydrates are the third most abundant class of breast milk components, and over 200 HMOs, comprised out of 5 monosaccharides (glucose, galactose, N-Acetyl-Glucosamine, fucose and sialic acid) have thus far been identified [23]. HMOs have recently moved into the spotlight of cognitive research due to its widespread effects on infant development and cognition [4,11,20]. There are three main families of HMOs; the non-fucosylated neutral HMOs, (e.g., Lacto-N-neoTetraose (LNnT)), the fucosylated HMOs (e.g., 2′Fucosyllactose (2′ -FL)) and the sialylated (SL) HMOs (e.g., 3 ′Sialyllactose (3-SL) and 6′ -Sialyllactose (6-SL)) [23,24]. Oligosaccharides are present in all mammalian milk [25]. However, what makes human milk unique compared to other mammalian milk is that it contains the largest diversity of complex oligosaccharides [25,26] and high concentrations of 2′ -FL. It should be noted that the presence of 2′ -FL is subject to large inter individual variation depending on the Lewis antigen blood group system of the mother, which encompasses two genes; the Lewis gene (Le gene or FUT-3 gene) and the Secretor gene (Se gene or FUT-2 gene) [27]. Depending on genetic expression, women are either defined as "secretors" (Se+), or "non secretors" (Se-), and Lewis positive (Le+) or Lewis negative (Le-) [23,27]. Both Secretor and Lewis genes are responsible for yielding fucosyltransferase-2 (FUT-2) and fucosyltransferase-3 (FUT-2) respectively, which append fucose to the core oligosaccharides. Depending on which of these FUT enzymes are active, different oligosaccharides will be created; as FUT2 expression results in the synthesis of 2′ -FL, while FUT3 expression has been associated with the formation of LNFP-II instead [28–30]. These polymorphisms essentially give rise to four major milk groups within the human population, as both genes can be active, inactive or either one of the two is active, hereby resulting in a variable HMO content in breast milk [29]. Around 60–72% of the maternal population are secretors, and the milk of these "secretor mothers" contains an overall higher concentration of HMOs in breastmilk as compared to non-secretors [23,27,31,32]. All in all, a large variability exists within the human population concerning the exact proportions of different HMOs [28]. Moreover, HMOs are also subject to dynamic changes within the same breastfeeding female, depending on factors such as circadian rhythm, lactation stage, maternal diet, and maternal genetic background [4,11,20,28–34].

Supplementation of infant formula with HMOs renders the composition and downstream effects of infant formula to become closer to those of breastmilk. One of the well-documented advantages of HMOs is its prebiotic role and the capacity to regulate the immune system in the periphery. HMOs can exert antimicrobial and antiviral effects by binding to pathogens which reach the mucosal surfaces in the gut or by directly binding to the gut epithelial receptors, effectively blocking the access of pathogens [11,20]. Experimental studies in infants showed enhancing effects on the immune response of additional 2′ -FL supplementation. Goehring and colleagues [35] observed that infants who were fed breastmilk or a 2′ -FL enriched formula had lower concentrations of plasma inflammatory cytokines (IL-1α, IL-1ß, IL-6, TNF-α) when compared to children fed the ordinary (non-enriched) infant formula [35]. Furthermore, ex vivo stimulation of peripheral blood mononuclear cells (PBMCs) yielded lower levels of TNF-α and IL-6 when infants were breastfed or were on a 2-'FL enriched diet. Enriching infant formula with 2′ -FL and LNnT also renders the gut microbiome composition and

its metabolites (propionate, butyrate and lactate) of formula-fed infants closer to that of breastfed infants [36]. It stands to reason that, if the supplementation of HMOs to infant formula produces immunological and health responses similar to those of breastfed infants, this may also partly account for cognitive outcomes [14]. Indeed, apart from HMOs involvement in immune functioning, a recent study by Berger and colleagues [37] reported that the amount of 2′ -FL, measured in mother's breast milk one month after birth, predicted improved cognitive outcomes in two-year-old children. Since it is known that alterations in the immune system impacts brain development and later life cognitive functioning [38], it is possible that the HMO mediated immune response provides a route via which HMOs could contribute to cognition. Thus, investigating how HMOs impact underlying neural mechanisms of their associated cognitive outcomes will provide valuable insight in HMOs' role in brain development and functioning.

While there have been correlational studies exploring the role of HMOs on development in humans, no direct human study has thus far investigated both immune and cognitive outcomes with HMO analysis in breast milk or upon HMO supplementation in infant formula. However, direct studies on the effects of HMOs and cognition have been undertaken in murine models and piglets. While there are obvious differences between species, several animal models have been used extensively in behavioural research due to their translational value in brain development and behaviour. The behavioural tasks used in animal models in probing various cognitive functions are well validated [39]. Moreover, since the life span of rodents in particular is relatively short, animal models allow the investigation of the most sensitive developmental period to HMO supplementation. In addition, behavioural studies in animals can be corroborated by more invasive measures in vivo, granting a live view on the underlying neurobiological processes. One method commonly used in rodent memory studies is electrophysiology. Long Term Potentiation (LTP) involves the strengthening of synapses in response to prior stimulation during memory formation and retrieval. This produces a long-lasting shift in synaptic strength and is therefore an important underlying mechanism of synaptic plasticity and memory [40]. Findings derived from preclinical work could prove to be informative and may serve as input to future longitudinal studies on the contribution of HMOs to the cognitive development of humans.

This review's aim is twofold. Firstly, it aims to summarize the effects of HMOs in animal research and their subsequent cognitive and electrophysiological outcomes. Special consideration is given to the type of HMO used (e.g., fucosylated (2′ -FL), neutral (LNnT) and sialylated (3′ -SL, 6′ -SL)), the age of the animals upon HMO administration, the used cognitive task complexity and the age of the animals during testing. Its second purpose is to provide additional interesting avenues for future research to explore. The search for relevant articles was conducted in Pubmed in the period of 1979 until August 2020, using a specialized search string comprised of both Mesh terms and key words in the title and abstract (Appendix A). This resulted in the inclusion of nine articles that contained (1) an animal model, (2) HMOs and (3) cognitive behavioural tests.

#### **2. Assessing the E**ff**ects of HMOs on Cognitive Measures in Animal Models**

Rodents and piglets are naturally curious and intelligent animals, which results in their frequent use as animal models for the assessment of cognition in a wide variety of behavioural tasks [41–45]. Behavioural tests are considered to be a valid, minimally invasive way to expose underlying cognitive processes, under the condition that the animal is capable of, and facilitated in, expressing such processes externally. In the context of HMO research, the focus has mainly been on memory and learning behaviour as cognitive capabilities. In the following sections, we will first graphically present an overview of the animal tests which investigated the consequences of HMOs on cognition. Subsequently, we present the main findings of the selected nine articles, grouped by the type of HMO (fucosylated or sialylated), in Table 1. Thereafter, the main results will be described, which is then followed by a discussion about the implications of the findings reported in the investigations.

The type of behavioural tests used to study the effects of HMOs on cognition make use of either the intrinsic rewarding value of an animal's natural curiosity in new exposures (Figure 1A,B,E) [41,42], the aversion to uncontrolled swimming without a platform to rest on (Figure 1D) [43] or the willingness to obtain an extrinsic reward like food or water (Figure 1C,F,G) [44,45]. Since animals prefer to be exposed to new items or environments to explore, the time spent to explore this new item or environment can be used as a measure for spatial or recognition memory. The willingness to obtain a food or water reward is commonly measured in operant conditioning tasks in either a skinner box or an Intellicage [44,45]. Operant conditioning tasks encompass associative learning paradigms, in which certain behaviour is reinforced via a reward or a punishment. In operant conditioning, different reinforcement schedules exist, such as the Fixed Ratio (FR) schedule, in which animals have to reach a certain criterion before they receive a reward. For example, an FR(4) schedule requires 4 correct responses from the animal in order for it to obtain a reward.

**Figure 1.** Summary of the behavioural tests used in the HMO studies. The type of animal placed inside the test (rodent or piglet) corresponds to the animal model used in the behavioural paradigms included in this review.

Overall, these cognitive tasks can be grouped by the level of complexity, as tasks that require a few trials are considered to be easier to perform than a task that requires weeks of training. In light of this, we have grouped the Y maze, T maze, Morris Water Maze (MWM) and the Novel Object Recognition

Test (NORT) as simple cognitive tests and the 8-arm radial maze and the operant tasks (Skinner box and Intellicage) as the complex cognitive tasks.

### **3. E**ff**ects of HMOs on Cognition in Mammals**

#### *3.1. Main Behavioural Findings*

Supplementing mammals with additional HMOs leads to beneficial cognitive outcomes under certain specific circumstances (Table 1, Figure 2). In general, both fucosylated and sialylated HMOs contribute to an improved memory performance and faster learning speed (tests described in Figure 1A–G) when tested in mature adulthood, irrespective of the age of administration of these HMOs (e.g., during infancy or adulthood) [46–54].

**Figure 2.** Graphical summary of behavioural tests results. The results have been grouped based on the type of HMO (Fucosylated versus Sialylated), animal model (rodents versus piglets) and the age of when the behavioural test has been performed. Infancy–young adulthood has been defined as the period ranging from PND1–6 months of age, while mature adulthood encompasses animals of 1 year old. Red crosses indicate that no significant differences were observed between the HMO and the control group, while green check marks indicate that positive effects due to HMO supplementation were reported. Details on the nature of such effects are summarized in Table 1.

#### 3.1.1. Simple Cognitive Tasks

When rodents performed spatial and recognition memory tests during adolescence and early adulthood, no effects of either fucosylated or sialylated HMOs, as assessed by the NORT (when tested 24 h after the acquisition phase), MWM and the Y maze, were reported. Contrary to the rodent studies, three piglet studies showed that supplementing HMOs during the lactation period resulted in improved spatial memory (T maze) in infancy [51] and object recognition (NORT) [52,53]. Supplementing only oligofructose or the combination of 2′FL and LNnT increased object recognition when piglets were tested one hour after the acquisition phase. When tested 48 h later, only the piglets who received a combination of either Bovine Milk Oligosaccharides (mostly neutral non fucosylated oligosaccharides) and 2′FL and LNnT [52] or Oligofructose and 2′ -FL [53] displayed long-term recognition memory. In mature adulthood (older than 1 year), rodent studies also found significant differences in both the Y maze and the NORT for both sialylated and fucosylated HMOs. However, the sialyllactose piglet study performed by Fleming and colleagues [54] yielded no results. In this study, they found no differences between the sialyllactose group and control group on the NORT performed during infancy.


#### **Table 1.**Summary behavioural studies.

#### **Table 1.** *Cont.*


NORT: Novel Object Recognition Test, MWM: Morris Water Maze, LTP: Long-Term Potentiation. BMO: Bovine Milk Oligosaccharide. When provided, strains of species have been included in the table. In all studies presented here, the HMOs were administered orally. All animals used in the studies were male, unless otherwise specified. When the experimental groups have not been detailed in the key results column, the reported n indicates the number of animals per experimental group of that study.

#### 3.1.2. Complex Cognitive Tasks

When considering the tasks that probe conditioning and learning capabilities and in which the cognitive difficulty could be varied, such as the 8-arm radial maze [50] and operant tasks [47–49], beneficial effects of HMO already surface at a young age in rats, mice and piglets alike. These effects also persist throughout adulthood. Perhaps the beneficial effects of HMOs become especially apparent upon increments on the cognitive load to meet the task demands.

#### *3.2. E*ff*ects of HMOs on Long Term Potentiation (LTP)*

The method of in vivo LTP induction in the studies listed here involved the implanting of stimulating electrodes on the Schaffer's collateral of the dorsal hippocampus and 2 to 4 recording electrodes in the stratum radiatum underneath CA1 [46–49]. A high frequency stimulation (200-Hz trains of pulses, 100 ms each and presented repeatedly with 1-min intervals) was delivered to the Schaffer's collateral and 30 min later the field excitatory post-synaptic potentials (fEPSPs) were recorded. Enhanced LTP responses are reported in all these studies [46–49], both after weaning and during adulthood, when animals were supplemented with fucosylated or sialylated HMOs.

#### **4. Discussion**

To the best of our knowledge, this review is the first to summarise the effects of fucosylated and sialylated HMOs on cognition and electrophysiological brain recordings in rodents and piglets. The effects of both types of HMOs uncovered in the reported investigations unequivocally point towards long lasting beneficial effects on cognition and memory, which is further supported by changes in the underlying physiological mechanisms as measured by LTP [46–49].

The majority of the reported animal studies, included in Table 1, revealed that HMOs enhance learning and memory. For the simple cognitive tasks, the effects of HMOs are not unequivocal, as differences are observed between the animal model used, task parameters, the dosage used and age of administration and testing. It should be noted that in the majority of the studies, the HMO dosage was comparable to concentrations found in human milk [28–31,33], and effects of HMO supplementation were already visible at these physiological relevant dosages.

In rodents, no significant effects on spatial memory or long-term recognition memory are reported when the animals' age ranges from juvenile to young adulthood. In piglets, HMOs are found to affect spatial memory and intermediate recognition memory but not long-term recognition memory when they were fed only HMOs during infancy. Inter species differences between rodents and piglets may help to explain why effects of HMO administration are visible in piglets but not in rodents when tested at a very young age. The third trimester in human gestation coincides with the first ten postnatal days of rat pups, while the neurodevelopmental trajectory and morphological properties of piglet brains are much more comparable to humans [41,55–57]. This complicates the comparison of the effects of oral delivery of HMOs between piglets and rodents. Differences reside in the immediate environment upon birth and the extent to which the brain and body have developed at that point, as neonatal rat pups would be more comparable to prenatal piglets in the final days before parturition, and there are no studies performed on the cognitive effects of HMOs on piglets in young adulthood. This interspecies difference in developmental stage upon birth and subsequent postnatal period might contribute to the heterogeneity in the findings between species on simple behavioural tests such as the NORT and the T maze.

Nevertheless, one cannot exclude the possibility that other factors than mere species differences may be at play, for example, the test parameters used in the studies. In the NORT of the rodent studies, the retention interval (time between acquisition phase and test phase) was 24 h, which is considered to be fairly long and is considered to be a measure of long-term recognition memory [42]. In the piglet studies, different retention intervals, ranging from 1 h (intermediate) to 48 h (long-term), were used. It is possible that similar enhancing effects of HMO administration on recognition memory (NORT)

reported by Fleming and colleagues [52,53] would have been found in juvenile rodents if the retention interval was 1 h instead of 24 h and if the rodents had been fed a similar combination of oligosaccharides as the piglets received. However, when probing such a long-term recognition memory of one-year old rodents, an improved recognition memory is observed in the HMO supplemented animals, together with improved spatial memory as measured by the Y maze. As long-term recognition memory was not observed in juvenile piglets and rodents when supplied with only one HMO but was observed in piglets when they were given a combination of oligosaccharides, this may not be a simple matter of species differences. Another explanation could be that within the developing brain, there are different processes at play when retrieving a newly consolidated memory (one hour later) versus an older memory (24–48 h later), which may require more resources, such as the combination of various types of oligosaccharides. Interestingly, when piglets were supplemented with a complex mixture of oligosaccharides (HMOs and BMOs or Oligofructose), they displayed an improved long-term recognition memory. Perhaps the effects when HMOs form combinations or are provided with other oligosaccharides are more potent and thus easier to discern than the effects of singular HMOs on memory.

Other factors such as gender and sample size could also contribute to the heterogeneity of the simple behavioural test findings, but it is uncertain to what extent these factors may have influenced the results. Only two sialyllactose studies (and no fucosyllactose study) used both males and females, one rodent study by Oliveros and colleagues [47] and one piglet study by Obelitz-Ryom and colleagues [51]. However, no separation based on gender was performed in the analysis. As studies on postnatal administration of compounds, such as the study by Shumake and colleagues [58] have demonstrated gender specific effects in rats, it stands to reason that early life HMO supplementation could produce gender specific outcomes. Nonetheless, when comparing the findings generated by Oliveros et al. [47] and Obelitz-Ryom et al. [51] with the exclusively male studies of the same species and HMO administered, the behavioural results remained very similar. Furthermore, the majority of the studies employed comparable sample sizes (*n* = 10–12 on average), and effects of HMOs on cognition were already reported in studies with the lower sample sizes. While potential effects of variation in sample size cannot be completely excluded, HMO supplementation already produces beneficial results in experiments with lower sample sizes. Therefore, the heterogeneity in findings between studies is more likely due to a combination of factors such as species and task parameters, as previously discussed.

When both piglets and rodents were tested on complex cognitive tasks from a young age onwards, HMOs exerted a beneficial effect on learning and memory. Therefore, it is possible that the HMOs effects become more apparent when cognitive load is increased, either due to task difficulty or due to aging. This may explain why the beneficial effects of HMOs are especially visible when the tasks are cognitively more strenuous, such is the case with the 8-arm radial maze or the operant tests, as increases in cognitive load make brain limitations more discernible.

While behavioural tests on learning and memory at a young age in general yielded mixed results, HMO supplementation did significantly improve LTP from a young age onwards. Interestingly, while in both young adult (2.5 months old) and mature adult (1 year old) just one HFS application was sufficient to induce LTP, very young rodents (6 weeks old) required a second high-frequency stimulation (HFS) to induce LTP. Nonetheless, HMO administration resulted in an enhanced LTP response in both younger and older rodents alike. It is possible that LTP might be a more sensitive measure to investigate the beneficial effects of HMOs on cognitive outcomes at a young age. Furthermore, under normal circumstances, the LTP response is reduced in older rats as a natural result of aging [46]. This natural reduction in LTP response was not encountered when the animals were supplemented with HMOs. On the contrary, supplementation with HMOs facilitated an enhanced LTP response. Because LTP is a measure of synaptic plasticity, it stands to reason that synaptic plasticity benefits from HMOs both in the short-term as in the long-term. Therefore, supplementation of HMOs, both sialylated and fucosylated, in infancy could have long-lasting protective effects on the molecular underpinnings of learning and memory.

It should be noted that these results have been gathered from only nine articles, which is the main limitation of the present review. Nevertheless, while there are a limited number of studies on the cognitive effects of HMO supplementation, the studies currently available show promising results of how HMOs could contribute to cognitive development. These findings call for further in-depth research on the cognitive effects of HMOs and to delineate their underlying mechanisms.

#### *Potential Underlying Mechanisms*

There are a few possible factors which could account for the cognition enhancing effects of HMOs in mammals.

In the case of sialylated HMOs, Polysialylated Neural Cell Adhesion Molecules (PSA-NCAM) could be upregulated. The PSA-NCAM complex is upregulated in newborn, immature neurons and growing fibre tracts during embryogenesis and has been linked to increased synaptic plasticity [59–62]. Within the adult brain, PSA-NCAM is expressed in brain regions with high rates of neural plasticity and neurogenesis, such as the olfactory bulb and the hippocampus [61]. Improved neural plasticity and the survival of newborn neurons contribute to cognition and memory [62]. Therefore, it is possible that sialylated HMOs are capable of influencing neurogenesis via upregulation of PSA-NCAM, which in turn contributes to the reported improvement in cognition. This suggestion is further supported by Oliveros and colleagues [47]. These authors found an increase in PSA-NCAM in 6-SL supplemented animals. However, the role of fucosylated HMOs in plasticity and neurogenesis is currently not well understood and requires further investigation.

A second possible factor is the improved immune functioning due to the supplementation of HMOs and their well-established role in the immune system. As mentioned in the introduction, immune factors also contribute to cognitive functioning [38], though there are multiple hypotheses on how this may occur. One hypothesis states that perinatal immune activation directly affects neurodevelopmental pathways necessary for learning and memory, which leads to reduced neurotransmitter function, a reduction in hippocampal presynaptic proteins and impaired LTP [38]. A second hypothesis postulates that early life immune activation indirectly determines the adult response to an infection with a pathogen, either via exaggerated pro-inflammatory cytokines or via a decrease in anti-inflammatory cytokines. This in turn could lead to downstream changes in cognition and neural function [38]. As HMOs are capable of regulating the neonatal cytokine response in the periphery [11,33,35,63], it is possible that they also exert their enhancing effects on cognition via the immune system.

A last possible factor through which HMOs may improve cognition involves the microbiome. HMOs contribute to the microbiome composition within the gut and therefore could interact with the brain via the resulting bacterial metabolites such as the Short Chain Fatty Acids [64]. As certain gut bacteria are specific for the utilization of sialylated HMOs and other bacteria for the fucosylated HMOs, a larger variety of HMOs may go hand in hand with a larger yield of specific gut bacteria capable of metabolizing these HMOs, and thus determining their subsequent metabolites [65]. Interactions between single HMOs and the microbiome have been previously reported by Tarr and colleagues [66]. They demonstrated that the administration of sialylated HMOs changed the microbial composition in the gut of mice, which in turn led to a reduction in anxiety-related behaviour and a maintenance of neurogenesis. The influence of the gut–brain axis has also been touched upon by Vazquez and colleagues [48], as they found that ablating the vagal nerve, which is part of the gut–brain axis, diminished the beneficial effects of orally supplied 2′ -FL on LTP. Similar to these results, Kuntz and colleagues examined the metabolic fate of 2′ -FL and found that 2′ -FL was not directly incorporated in the brain but required an intact gut microbiome for the generation of fucose metabolites, which are subsequently taken up into the systemic circulation and organs [67]. In addition, it is possible that combinational HMOs may generate better effects than alone. This idea has already been demonstrated at the level of the growth and function of gut bacteria [68,69]. Different HMOs are processed by different bacteria, which contain either sialidases or fucosidases to cleave sia and fuc of the carbohydrates [65]. In turn, another group of bacteria can feed on the HMOs once the fuc and sia moieties are removed. These bacterial interactions, which depend on the HMOs present in the gut, may exert downstream effects on memory and cognition via the gut–brain axis. In light of potential downstream effects of the microbiome on behaviour, environmental housing conditions which affect the microbiome should also be considered [70] in this context, although it is uncertain to what extent the microbiotic variations due to husbandry may have influenced the effects of HMO supplementation on subsequent behaviour. Finally, another important factor to consider in the context of the microbiome is gender specific effects. While infant sex is reported to be largely unrelated to the HMO composition within human breastmilk [31], another study by Moossavi and colleagues [71] found that the milk microbiota vary depending on infant sex. This could potentially be attributed to cross interactions with the gut microbiome of the infant, as gender differences have been reported there [71]. As HMOs interact with both the milk and the gut microbiome [72,73], it is therefore possible that sex-dependent variations could lead to differential cognitive outcomes of HMO supplementation.

#### **5. Conclusions**

The observation that HMOs are capable of enhancing cognition has initiated the search for a better mechanistic understanding of its functioning. Nonetheless, there are still several outstanding questions on the relationship between HMO and neonatal brain development, which warrant further investigation. An important aspect that needs to be addressed is the apparent age-related differences when assessing various cognitive tests. This point illustrates one of the current issues on HMO research in animals, as the tools currently used may not be sensitive enough to fully explore the range to which HMOs may affect brain development and cognition. Thus, one of the more complex tools could be the use of challenging operant tasks, such as the Trial Unique Delayed Non-Matching to Location (TUNL) measuring spatial memory and pattern separation [74], the 5-Choice Serial Reaction Time Task measuring attention and motor impulsivity [75], or delayed reinforcement tasks measuring choice impulsivity [76], ideally performed in the animal's home cage. The difficulty of such tasks can be varied and may thus be more suited to test cognitive functioning in young animals, as at a young age, only effects of HMOs were found in difficult tasks.

Another important issue is that due to the large variability between the experimental design and methods used across studies, comparing the effects of different HMOs between studies is difficult. Such variability includes the age of testing, the tests and experimental parameters, the HMO components used, the gender of the animals, variation in sample sizes, the environmental conditions and the variation in (neuro)developmental stage during which the animals were supplemented the HMOs. These limitations call for a larger, unified study in which the effects of different HMOs on complex cognitive functioning are systematically compared, when administered both independently and as in conjunction. In such a unified study, all these factors can be accounted for, enabling a systematic comparison.

A last important issue is that most HMO studies so far have focused on singular HMOs, with the exception of the two most recent studies performed by Fleming in 2020. The focus on singular HMOs is a limitation because it does not reflect a naturalistic situation where maternal milk provides a combination of different HMOs [77]. Therefore, considering the interactions of HMOs when supplemented in combination would provide valuable insights on the influence of the gut microbiome and its downstream effects on cognition and development.

While research on the cognitive implications of HMOs is still in its infancy, the early findings reporting its long-lasting beneficial effects on memory and cognition are promising. Further studies on the exact molecular mechanisms, ranging from immune functioning to neuroplasticity and the microbiome will prove to be useful in deepening our understanding of how HMOs and their interactions contribute to cognition and development.

**Author Contributions:** Conceptualization, S.D., and J.R.H.; writing—original draft preparation, S.D.; re-writing, S.D., review and editing, S.D., J.R.H., M.S. and W.W.; Table and Figure creation, S.D.; supervision, J.R.H. and M.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** W.W. is an employee of Inner Mongolia Yili Industrial Group, Co., Ltd. The other authors declare no conflicts of interest.

#### **Appendix A**

The following search string, comprised of both Mesh terms and key words in the title and abstract was composed and entered in Pubmed:

(Oligosaccharides[MeSH Terms] AND milk, human[MeSH Terms] OR oligosaccharide\*[Title/ Abstract] OR HMO\*[Title/Abstract]) AND (learning[MeSH Terms] OR cognition[MeSH Terms] OR memory[Title/Abstract] OR cognition[Title/Abstract] OR behavior[Title/Abstract] OR behaviour[Title /Abstract]) AND ("sialic acids" [MeSH Terms] OR sialyl\*[Title/Abstract] OR fucosyl\*[Title/Abstract] OR fucose[MeSH Terms] OR lacto-N-\*[Title/Abstract] OR LNnT[Title/Abstract])

This search string enables the detection of articles that contain information regarding HMOs, their cognitive outcomes and specific HMO components. Using the above search strategy, 108 articles were found. Further exclusion criteria were articles that did not pertain to cognition or its associated processes (learning and memory) and that did not make use of administered HMOs. This finally resulted in 9 articles.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **The Role of the Gut Microbiota in the Development and Progression of Major Depressive and Bipolar Disorder**

**Tom Knuesel and M. Hasan Mohajeri \***

Department of Anatomy, University of Zurich, Winterthurerstrasse 190, 8057 Zürich, Switzerland; tom.knuesel@uzh.ch

**\*** Correspondence: mhasan.mohajeri@uzh.ch; Tel.: +41-79-938-1203

**Abstract:** A growing number of studies in rodents indicate a connection between the intestinal microbiota and the brain, but comprehensive human data is scarce. Here, we systematically reviewed human studies examining the connection between the intestinal microbiota and major depressive and bipolar disorder. In this review we discuss various changes in bacterial abundance, particularly on low taxonomic levels, in terms of a connection with the pathophysiology of major depressive and bipolar disorder, their use as a diagnostic and treatment response parameter, their healthpromoting potential, as well as novel adjunctive treatment options. The diversity of the intestinal microbiota is mostly decreased in depressed subjects. A consistent elevation of phylum Actinobacteria, family Bifidobacteriaceae, and genus *Bacteroides*, and a reduction of family Ruminococcaceae, genus *Faecalibacterium*, and genus *Roseburia* was reported. Probiotics containing *Bifidobacterium* and/or *Lactobacillus* spp. seemed to improve depressive symptoms, and novel approaches with different probiotics and synbiotics showed promising results. Comparing twin studies, we report here that already with an elevated risk of developing depression, microbial changes towards a "depression-like" microbiota were found. Overall, these findings highlight the importance of the microbiota and the necessity for a better understanding of its changes contributing to depressive symptoms, potentially leading to new approaches to alleviate depressive symptoms via alterations of the gut microbiota.

**Keywords:** depression; affective disorder; gut-brain-axis; bacteria; probiotics; therapy; treatment

#### **1. Introduction**

Psychiatric disorders belong to the world's most disabling diseases, particularly major depressive disorder (MDD, unipolar disorder) and bipolar disorder (BD). Approximately 4.4% of the world's population is affected by depression. According to the World Health Organization, depression is the largest contributor to global disability and "non-fatal health loss", as well as the major contributor to suicide deaths [1]. Patients with MDD show typical symptoms of sadness, loss of interest and pleasure, feelings of low self-worth, guilt and tiredness, disturbed sleep, and poor concentration. BD is characterized by episodes of depression and mania, separated by episodes of normal mood. Mania includes elevated mood, increased energy and activity, pressure to speech, and decreased need to sleep [1].

In twin and family studies, heritability rates, defined as genetic factors contributing to the occurrence of a certain disease, were found to be moderate in MDD [2], and high in BD [3,4]. Despite significant advances, the pathogenesis of MDD and BD is still not fully understood. The diagnosis is based only on clinical symptoms, and a high rate of treatment resistance is observed [5]. Poverty, unemployment, severe life events, physical illness, and the consumption of alcohol and drugs are risk factors, but anyone can be affected by depression [1]. Especially functional gastrointestinal disorders (FGID) like irritable bowel syndrome (IBS) are often associated with depression, with the co-occurrence estimated at 30% [6]. Altered neurotransmission, changes in the hypothalamic-pituitary-adrenal axis (HPA axis), chronic low-grade inflammation, reduced neuroplasticity, and neuronal

**Citation:** Knuesel, T.; Mohajeri, M.H. The Role of the Gut Microbiota in the Development and Progression of Major Depressive and Bipolar Disorder. *Nutrients* **2022**, *14*, 37. https://doi.org/10.3390/nu14010037

Academic Editor: Giovanna Muscogiuri

Received: 30 November 2021 Accepted: 20 December 2021 Published: 23 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

network dysfunction probably contribute to the pathogenesis of depression [7]. IBS pathogenesis shares several of these changes, indicating a multifactorial association between both diseases [8]. Additional evidence suggests a connection between depression, increased gut wall permeability, and bacterial translocation, resulting in increased immune activation and inflammation, with the intestinal microbiota being an important contributor [9,10].

The human gut microbiota consists of an estimated number of 3.8 × 10<sup>13</sup> (38 trillion) bacteria, containing slightly more bacteria than cells of the human body (approximately 3.0 × 1013), and by far more genes than its human host [11]. In addition, after the brain, the human gut contains the second greatest number of neurons. Heritability rates of the gut microbiota in humans were estimated between 1.9% and 8.1% [12]. A disturbed intestinal microbiota, often associated with reduced diversity, was found in a variety of diseases, including hypertension [13], obesity [14], gastrointestinal disorders (such as inflammatory bowel disease (IBD) [15,16], and IBS [17]), brain disorders (such as Alzheimer's disease [18], Parkinson's disease [19], autism spectrum disorder [20], and attention-deficit/hyperactivity disorder [21]), autoimmune diseases [22], as well as some types of cancer (for example colorectal cancer [23]). It was even suggested that a disturbed intestinal microbiota in obese patients may be a further reason for increased coronavirus disease 2019 (COVID-19) severity [24].

The intestinal microbiota can interfere with the HPA axis. Stress-induced stimulation of the HPA axis leads to elevated adrenocorticotrophin (ACTH) release and therefore results in a higher glucocorticoid excretion. In restraint-stressed germ-free mice, elevated ACTH and corticosterone (a glucocorticoid) levels were found, compared to specific pathogen-free mice [25], showing a direct connection between the HPA axis and the microbiota. The HPA axis can also be influenced through metabolites produced by the intestinal bacteria, like cytokines and prostaglandins, leading to exaggerated or attenuated stress response [26]. The gut microbiota can break down otherwise indigestible food substances and produce micronutrients [27], short-chain fatty acids (SCFAs) [28], neurotransmitters such as gammaaminobutyric acid (GABA) [29], and brain active non-SCFA metabolites [18]. Acetate, propionate, and butyrate are the most abundant SCFAs in the human intestine [30]. They can influence emotion, cognition, and the immune system. In particular, a correlation between higher depression scores and lower levels of acetate and propionate was found in women, while sodium butyrate reversed depressive and manic symptoms in mice and was suggested as a mood stabilizer in humans [31,32]. Bacterial metabolites can translocate out of the gut and interact with the HPA axis, with the immune system, and with vagal afferents, leading to an exaggerated (or attenuated) HPA response and consequently to a modulation of the immune system [26,33,34]. Further research reported a systemic chronic low-grade inflammation in mice models, as well as in a significant proportion of depressed subjects, suggesting the presence of a mucosal dysfunction in depressed individuals, leading to an elevated translocation of intestinal bacteria into the circulation [35–37]. Consequently, an increased antibody response against lipopolysaccharides (LPS) from gram-negative bacteria is induced in diseased individuals [9]. In mice, intraperitoneally injected LPS caused a depressive-like behavior, and a following treatment with sodium butyrate ameliorated these changes, underlining the negative influence of translocated bacteria and LPS, as well as the positive influence of butyrate on the depression pathophysiology [35].

Several possible connections between the intestinal microbiota and depression are currently being discussed. The gut microbiota is considered to be under-explored, and its detailed investigation is needed for revealing specific associations. Most studies examining a possible connection between the gut microbiota and depression are conducted in rodents, while human research is still lagging. Hence, we systematically reviewed the connection between the human intestinal microbiota and major depressive and bipolar disorder, intending to analyze which bacteria could possibly influence depression or vice versa, and which bacteria future studies should primarily focus on.

#### **2. Materials and Methods**

The main question of this review was whether there is a connection between the intestinal microbiota and major depressive and bipolar disorder in human subjects. Does the intestinal microbiota influence the development, severity, and remission of affective disorder? The databases Scopus and PubMed were searched until 1 May 2020, with the following MeSH and search terms: "microbiota", "microbiome", "depression", "depressive", "bipolar disorder".

Additional inclusion criteria were as follows:


We focused on bacterial taxa and therefore excluded results regarding fungi, archaea, and viruses. Studies investigating microbiomes other than the intestinal microbiota were also excluded. We included all studies related to MDD, BD, and the intestinal microbiota, leading to the high heterogeneity of the reports, but providing a comprehensive overview of published data on this topic.

Twelve articles were excluded after full-text assessment due to not focusing on depression, depressive symptoms (*n* = 9) or the intestinal microbiota (*n* = 3).

A total of 57 studies were included in this review (Figure 1), most of which were published between 2016 and 2020, demonstrating a rapidly growing interest in this topic in recent years.

**Figure 1.** Methodical approach of our review due to PRISMA criteria [38].

#### **3. Results**

#### *3.1. Diversity*

Microbial diversity can be specified as alpha-diversity and beta-diversity. Alphadiversity describes the species richness and evenness (inequality of the relative abundance) within a sample. In the reviewed studies it was most often determined by using the Shannon index, but several measures are common for richness and evenness estimation, as the ACE-, Chao1-, and Simpson index, phylogenetic diversity, and the number of observed species [39]. Beta-diversity describes the difference between multiple samples and is mostly analyzed by using unweighted/weighted UniFrac distances and Bray-Curtis dissimilarity [40]. Additionally, PLS-DA (partial least squares discriminant analysis) was used to detect microbial patterns that separate depressed subjects from healthy controls (HC) [41].

Apart from Jiang et al. [42] who reported an increased alpha-diversity in active-MDD subjects, all alpha-diversity indices, and other measures were consistently reported to be equal or reduced in depressed subjects among the other studies. For example, four studies reported a negative correlation between the Shannon index and depression, while 13 reported no correlation. Similar results were found concerning the other abovementioned indices and measures for "within-sample" alpha-diversity (Table 1).

Regarding "between-samples" beta-diversity changes, a correlation with depression could be found in most studies. Both studies comparing MDD and BD subjects to HC using Bray-Curtis dissimilarity found a significant difference [43,44]. Regarding unweighted and weighted UniFrac distances, studies reported contradictory results, and no final statement can be made as to whether depressed subjects show different UniFrac distances compared to HC (Table 1).

Using PLS-DA, all four studies found significant differences between the depressed and the HC group. In addition, with a PLS-DA model, Li et al. found significant correlations between microbial and mood changes in healthy adults over time [45].

**Table 1.** A selection of the most used diversity indices and measures, showing an unchanged or lower microbial diversity in depressed individuals.


"UniFrac" includes weighted and/or unweighted UniFrac distances. "↓" shows a significantly reduced diversity in diseased subjects compared to controls or an inverse correlation with more severe symptoms, while "↑" indicates a significantly increased diversity or a positive correlation with more severe symptoms. "=" demonstrates no significant difference, while "sign." shows a significant difference. Empty cells symbolize that no results were reported. Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; C, control group; P, psychiatric subjects; D, depression in general; IBS, irritable bowel syndrome; a, active disease group; r, response group; pm, psychiatric measures; OTU, operational taxonomic unit; \*, only showing a trend, due to small sample size.

Overall, the microbial diversity of depressed individuals was reported unchanged or reduced, compared to the general population. These findings support the hypothesis that the intestinal microbiota is connected with the development, preservation, and remission of depression. With a better understanding of this link, depressive symptoms could potentially be positively influenced by specific alterations of the microbiota. Our analysis also shows that a reduced diversity is not present in all depressed subjects, and therefore cannot be used as a reliable diagnosis parameter. However, using the diversity change over time as a treatment response and prognosis parameter could be possible, but whether this is clinically feasible remains to be proven.

#### *3.2. Phylum*

On the phylum level, Bacteroidetes phylum was reported to be reduced in depressed subjects in multiple studies, but other studies reported opposite results (Table 2). Important to mention is that Chen et al. reported an elevation of Bacteroidetes in young MDD individuals, while in middle-aged MDD individuals Bacteroidetes were reduced, compared to age-matched controls [49]. On the one hand, comparing the studies which reported an elevation of Bacteroidetes in depressed patients, Liu et al. [44] only included MDD subjects between age 18 and 25, Hu et al. [66] used a young BD group (mean age 24 years), with an older, not age-matched control group, and Jiang et al. [42] only included MDD patients aged 40 or younger. On the other hand, studies reporting a reduction of Bacteroidetes correlating with depression were mostly conducted with older subjects. For example, in the study by Lai et al. [43], MDD individuals were between 32 and 52 years old, in Rong et al. [50] the mean age of all groups was between 38 and 42 years, and in Chen et al. [67] MDD subjects were 44 years old on average. Summarized, strong evidence was found that in young patients suffering from affective disorder, phylum Bacteroidetes is elevated, while in middle-aged patients these bacteria are reduced, compared to age-matched HC. This could point towards different causes of depression or a different manifestation of depressive behavior with age, leading to a different microbiota in depressed subjects.

The phylum Firmicutes was reported to alter in both directions (Table 2). However, most studies reporting an increase of Firmicutes also found a reduction of Bacteroidetes, or vice versa. Above mentioned changes due to different age ranges can also be applied to the phylum Firmicutes. Chen et al. [49] found lower Firmicutes mainly in the young MDD group, and Jiang et al. [42], Liu et al. [44], and Hu et al. [66] all studied young MDD or BD subjects and reported a reduced abundance of Firmicutes in the depressed group. Studies with older (middle-aged) subjects rather reported an elevation of Firmicutes abundance (Table 2).

Concerning the phylum Proteobacteria, results were highly controversial. Zheng et al. [46] and Rong et al. [50] both reported an elevation in BD, but not in MDD subjects, suggesting a potential difference between BD and MDD. In contrast, Hu et al. [66] found reduced Proteobacteria in untreated, compared to treated BD subjects, but not compared to HC. A recent review reported a lower abundance of phylum Proteobacteria in healthy subjects, while an elevated abundance was associated with a variety of diseases, including IBD, metabolic disorder, or malnutrition [68]. However, a link between phylum Proteobacteria and depression could not be shown, and neither age-dependent changes nor a difference between BD and MDD subjects could be further supported in this review.

The phylum Actinobacteria was consistently found to be elevated in MDD or BD individuals. Nine studies reported an elevation, while none found a decrease of Actinobacteria in patients with affective disorder (Table 2). Therefore, strong evidence for a close connection between an elevation of phylum Actinobacteria and depression was found, and more attention should be paid towards this phylum, looking for possible causes and consequences of an increase in Actinobacteria abundance.

Summarized, age seemed to firmly influence bacterial abundance. While in youngaged patients Bacteroidetes were elevated and Firmicutes reduced, in middle-aged subjects a reduction of Bacteroidetes and an elevation of Firmicutes was reported, compared to agematched HC, whereas Actinobacteria was consistently elevated regardless of age. However, on lower taxonomic levels correlations could show opposite directions even for closely related bacteria, suggesting that the abundance of a specific phylum is not as decisive as the abundance of certain bacteria on lower taxonomic levels, i.e., on the genus or species level.



Elevated Bacteroidetes and reduced Firmicutes were found in young-aged, depressed subjects, while in middleaged those phyla showed opposite correlations. Actinobacteria was consistently increased. "↓" and green symbolizes a significant reduction in diseased subjects or inverse correlation with more severe symptoms, while "↑" and orange shows a significant elevation or positive correlation with more severe symptoms. Grey symbolizes only a trend, while empty cells symbolize that no significant results were reported. Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; C, control group; D, depression in general; non-D, non-depressed subjects; Di, distressed subjects; IBS, irritable bowel syndrome; P, psychiatric subjects; a, active disorder group; r, response group; \*, only showing an insignificant trend.

#### *3.3. Bacteroidetes*

The phylum Bacteroidetes is the most dominant in the human gut [70]. It contains four important families, namely Bacteroidaceae, Tannerellaceae, Prevotellaceae, and Rikenellaceae.

Most studies reported a correlation between depressive symptoms and the family Bacteroidaceae or genus *Bacteroides*, the most abundant family and genus of the intestinal microbiota [70]. Especially genus *Bacteroides* was repeatedly associated with affective disorder, high anhedonia, and negative mood (Table 3). In general, *Bacteroides* were found to be negatively associated with inflammation [50,71], to contribute to the gut colonization resistance (the resistance against colonization of enteric pathogens), and to produce SCFAs, mostly acetate and propionate, which are important for gut homeostasis [72,73]. *Bacteroides* are known as starch degraders and they potentially cross-feed other species, like *Eubacterium ramulus*, which in turn can produce beneficial molecules like butyrate, and therefore reduce gut hyperpermeability by increased expression of tight-junctions [72]. Even though genus *Bacteroides* was found to provide beneficial effects on the human host, the family Bacteroidaceae and genus *Bacteroides* were repeatedly found to be elevated in depressed subjects. While some of the reviewed studies were conducted only with few individuals, Cheng et al. included thousands of subjects, strongly supporting this correlation [74]. These findings may hint at a compensatory mechanism and suggest that the alteration in the abundance of a certain bacteria may not necessarily have negative health effects. A higher taxonomic resolution could lead to more precise information about these correlations.


**Table 3.** Different abundance of genus Bacteroides.

Most studies reported an elevation correlating with affective disorder and depressive symptoms. "↓" and green symbolizes a significant reduction in diseased subjects or inverse correlation with more severe symptoms, while "↑" and orange shows a significant elevation or positive correlation with more severe symptoms, and grey symbolizes alterations in both directions or not evaluable. Brackets include additional information about the reported correlation (which bacteria showed a correlation or in which subgroup of subjects a correlation was found). Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; D, depression in general; IBS, irritable bowel syndrome; B-P group, Bacteroides-Prevotella group; m, a correlation only in male subjects; a, active disease group; r, response group; OTU, operational taxonomic unit within genus Bacteroides.

Genus *Parabacteroides* of the Tannerellaceae family tended to be elevated in depressed subjects, but two studies reported contrary results [60,69]. These two studies used a small sample size of depressed subjects (*n* = 15), and reported a reduction in non-depressed participants to correlate with anxiety and DASS-42 (depression anxiety and stress scales) scores, but not directly with depression, respectively [60,69]. Investigating the three studies which compared MDD or BD subjects with HC, all three reported an elevation of *Parabacteroides* correlating with depression [42,51,66]. *Parabacteroides* produce SCFAs, especially acetate, and can reduce neutrophils in the blood [76]. Even though they have health-promoting effects, they tended to be more abundant in individuals with affective disorders. Therefore, as with the family Bacteroidaceae, an elevation of *Parabacteroides* could be a compensatory mechanism, rather than unfavorably influencing the host's mood.

The abundance of the Prevotellaceae family altered in both directions in depressed subjects, with no tendency overall. Worth mentioning is that Chen et al. found a reduction of Prevotellaceae in middle-aged MDD, compared to young-aged MDD individuals [49]. This goes in line with our general findings, with two studies reporting a reduced Prevotellaceae abundance in middle-aged MDD subjects (mean age 45.8 and 43.9 years, respectively) [51,67]. On genus level, *Prevotella* inversely correlated with depression, lower mood, or lower quality of life in four studies [42,45,51,77], while others reported a positive correlation [60,62]. Therefore, the suggestion of Lin et al. [62], to use changes of *Prevotella* and *Klebsiella* for laboratory diagnosis and treatment evaluation in MDD, could not be further supported regarding *Prevotella* changes, because the results showed no clear tendency and additional studies even found an opposite correlation. Concerning *Klebsiella* changes, more research is needed to be able to draw a conclusion (further information in the paragraph "Proteobacteria"). Interestingly, two studies conducted with IBS subjects reported an elevation of Prevotellaceae, as well as an elevation of *Prevotella* and *Paraprevotella* to correlate with depressive symptoms, indicating a potential correlation with IBS and comorbid depression [55,60]. However, due to small sample sizes and different study designs of these two studies, additional research focusing on bacterial alterations in IBS subjects is required.

Within the family Rikenellaceae, results showed an elevation of genus *Alistipes* or operational taxonomic units (OTUs) within this genus to correlate with depression [41,42,63]. While *Alistipes* seemed to attenuate the severity of colitis via attenuating the expression of anti-inflammatory cytokines in mice, this genus was found to be increased in stressed mice, as well as in patients suffering from chronic fatigue syndrome. It is proposed to decrease

serotonin concentration and therefore negatively influence the gut-brain axis, which is in line with our conclusion that an elevated *Alistipes* abundance is associated with unfavorable health effects and potentially promotes the pathogenesis of depression [78]. More studies are needed to further investigate the influence of the Rikenellaceae family on depression.

#### *3.4. Firmicutes*

The phylum Firmicutes is the second most abundant phylum in the human intestinal microbiome [70]. It was also the most changed, as well as the most discussed phylum within the reviewed studies.

Class Bacilli includes two important families, the families of Lactobacillaceae and Streptococcaceae. *Lactobacillus* bacteria are widely known for their beneficial health effects and their use as probiotics. Despite several studies reporting a higher abundance of *Lactobacillus* to be associated with diverse positive factors like sleep and self-judgment, no direct correlation of *Lactobacillus* abundance and affective disorder could be identified [79,80]. Family Streptococcaceae and genus *Streptococcus* showed a positive correlation with depression and lower quality of life scores. For example, beta-hemolytic Streptococcus Group A infections are known to potentially cause Pediatric Autoimmune Neuropsychiatric Disorders (Associated with Streptococcal Infections, "PANDAS"), which are associated with alterations in the gut microbiome and the nervous system [81]. Although not fully understood, it demonstrates that *Streptococcus* infections can lead to an autoimmune response, severe brain alterations, disturbed neurotransmitters, and can cause psychiatric symptoms like obsessive-compulsive disorder, tics, anxiety, and sometimes even depression [82]. Of the included studies, four reported an elevation of Streptococcaceae (or OTUs within this family) and *Streptococcus* in MDD and BD [50,51,62,63]. Additionally, investigating a large cohort, Valles-Colomer et al. found a negative association between *Streptococcus* and body pain, but no direct association with depression, maybe due to the cohort representing the general population and not being limited to specifically MDD or BD subjects [77]. Hence, there is a need for more studies on the Streptococcaceae family and *Streptococcus* genus regarding their influence on mental health, with potential for novel therapeutic approaches.

Concerning the class Clostridia, a reduction of class Clostridia or order Clostridiales seemed to correlate with worse health and depressive symptoms. However, on lower taxonomic levels, probably due to the diversity of the family Clostridiaceae [83] and an insufficient number of well-controlled studies, no clear association could be identified between Clostridiaceae or *Clostridium* and depression, and studies reported ambiguous findings. There is a need for in-depth studies at high taxonomic resolution to further investigate a potential connection between certain genera or species within the family Clostridiaceae and affective disorder. In addition, an antidepressant therapy with adjunctive probiotic *Clostridium butyricum MIYAIRI 588* showed a high response rate with significant improvement of depressive symptoms in treatment-resistant MDD subjects (further information in the paragraph "human interventional trials in depression") [84]. Family Christensenellaceae and Christensenellaceae R-7 group were reported to be less abundant in subjects with affective disorder and to inversely correlate with more severe symptoms and higher anhedonia in three studies, while none reported opposite results [44,47,53]. Family Christensenellaceae has been shown to produce acetate and butyrate, and was negatively associated with visceral fat mass [85,86]. Even though, to our knowledge, not much is known about this family so far, a higher abundance of family Christensenellaceae is related to beneficial health effects, while depression correlates with a reduction of this family.

Family Peptostreptococcaceae or genus *Peptostreptococcus* were associated with depression [51,58] and anxiety [69] in three studies. In general, while some studies found a connection between *Peptostreptococcus* species and colorectal cancer [87], others proposed a beneficial effect via the production of indoleacrylic acid (a metabolite of tryptophan), improving the intestinal epithelial barrier function, as well as suppressing inflammatory response [88]. Despite these potentially beneficial effects, a rather negative correlation between *Peptostreptococcus* and depression was reported in the reviewed studies. But with

only two studies finding a significantly altered abundance in small cohorts of depressed individuals, while none of the studies with more subjects reported similar results, an important association with depression is unlikely.

Family Eubacteriaceae and especially species *E. rectale* tended to be more abundant in healthy subjects and increased after antipsychotic BD treatment with quetiapine [44,47,57,75]. *Eubacterium* spp. can produce propionate and butyrate, and therefore suppress inflammation, enhance the intestinal barrier integrity, and thereby benefit the host's health [89]. This is in line with our findings, where Eubacteriaceae correlated with better general health. But the evidence in terms of an association with depression remains scarce.

Family Lachnospiraceae is the second most abundant family in the human gut [70]. On the family level, studies did not show a clear connection between Lachnospiraceae and depression, with Lachnospiraceae being altered in both directions in depressed individuals. However, on the genus level, several inverse correlations with depression were found. Genus *Coprococcus* and OTUs within this genus were found to be less abundant in depressed subjects and to correlate with higher quality of life [46,48,58,60,63,66,77]. One study found specifically *C. catus* to be less abundant in subjects with more severe depressive symptoms and to positively correlate with remission [48]. Genus *Coprococcus* is known for its butyrate production [90], and previous research found a reduced *Coprococcus* abundance in several diseases, like inflammatory bowel disease [16], colorectal cancer [91], and preeclampsia [90]. According to Zhang et al., *Coprococcus* abundance can be increased by omega-3 polyunsaturated fatty acids (PUFAs), while lower levels of omega-3 PUFAs were found in depressed subjects [92,93]. Therefore, a connection between genus *Coprococcus*, PUFAs (especially omega-3 PUFAs), and depression is imaginable, emphasizing the importance of a healthy diet and its influence on the intestinal microbiota and depression. The abundance of genus *Fusicatenibacter* or unclassified species within this genus were reduced in depressed subjects and associated with a higher quality of life in three studies, indicating a slightly beneficial effect of these bacteria [44,55,77]. Due to contradictory results concerning genus *Blautia*, no final statement could be made regarding a link of genus *Blautia* with depression. Genus *Roseburia* or OTUs within this genus were mostly reported to be reduced in subjects with depressive symptoms and correlated with remission and positive mood (Table 4). Research showed that *R. intestinalis* suppressed inflammation and promoted anti-inflammatory cytokines in a colitis mouse model, and found *Roseburia*, together with *Faecalibacterium*, to be one of the most abundant known butyrate-producing bacteria in the human gut [94,95]. This is in line with our findings of *Roseburia* being reduced in depressed individuals. Therefore, an increase of bacteria belonging to the genera *Roseburia* or *Coprococcus* may provide beneficial physical and mental health effects, and further investigation is needed as to whether this effect can be used in the treatment of affective disorder.

Concerning the family Ruminococcaceae, most studies found a higher abundance in healthier subjects (Table 5). Interestingly, the abundance of family Ruminococcaceae was associated with remission, but several taxa within Ruminococcaceae positively correlated with symptom severity of psychiatric subjects [48]. Additionally, both increased and decreased OTUs within this family were found in MDD patients, underlining the differences of bacterial abundance on low taxonomic levels [49,57]. On genus and species level, on the one hand, multiple studies reported a higher abundance of genus *Oscillibacter* in depressed subjects [41–43,50]. Genus *Oscillibacter* is suggested to be elevated rather as a result of depression, due to its potential to metabolize proteins [96]. Underlining this, in MDD subjects, disturbed bacterial proteins were found, which are involved especially in metabolic pathways related to amino acid metabolism [67]. On the other hand, genus *Ruminococcus*, *Gemmiger*, and especially *Faecalibacterium* were more abundant in healthier subjects in the majority of the reviewed studies (Table 5). *Faecalibacterium* is known for its butyrate production, anti-inflammatory potential, and intestinal barrier function improvement, and was suggested as a probiotic for IBD, gut dysfunction, and low-grade inflammation treatment [95,97,98]. In the reviewed studies, a negative correlation with depressive symptoms and a positive correlation with remission and higher quality of life

was reported, highlighting the potential of probiotic *Faecalibacterium* as a novel treatment option, and their abundance as a parameter for diagnosis or treatment response. Even though several studies reported opposite correlations, contradictory results can mostly be explained by very small sample sizes, age and sex differences, and by not unexceptionally used false discovery rate. In conclusion, the family Ruminococcaceae is a perfect example that even closely related bacteria can show an altered abundance in opposite directions. Even though on the family level, most of the studies reported a higher abundance in healthier subjects, on lower taxonomic levels bacterial alterations were found in both directions. Similar to other families, a higher taxonomic resolution of this family and its genera is needed for a more specific examination of these bacteria and their interaction with the host, with especially genus *Faecalibacterium* showing a close negative association with depression.


**Table 4.** Different abundance of genus Roseburia.

Most studies reported a reduction correlating with affective disorder and negative mood. "↓" and green symbolizes a significant reduction in diseased subjects or inverse correlation with more severe symptoms, while "↑" and orange shows a significant elevation or positive correlation with more severe symptoms, and grey symbolizes only a trend. Brackets include additional information about the reported correlation (which bacteria showed a correlation or in which subgroup of patients a correlation was found). Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; D, depression in general; f, a correlation only in female subjects; a, active disorder; OTU, operational taxonomic unit within genus Roseburia; \*, only negatively correlating with symptom severity, but not significantly correlating with BD compared to healthy controls.

**Table 5.** Different abundance of family Ruminococcaceae and two of its members, genus Ruminococcus and Faecalibacterium.




significant reduction in diseased subjects or inverse correlation with more severe symptoms, while "↑" and orange shows a significant elevation or positive correlation with more severe symptoms, and grey symbolizes alterations in both directions. Brackets include additional information about the reported correlation (which bacteria showed a correlation or in which subgroup of patients a correlation was found). Empty cells symbolize that no significant results were reported. Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; D, depression in general; P, psychiatric subjects; DASS, depression anxiety stress scales; QoL, quality of life; f, a correlation only in female subjects; a, active disorder; OTU, operational taxonomic unit.

Additionally, an elevation of genus *Flavonifractor* was associated with depression, symptom severity, or worse physical functioning, with no contradictory results [42,44,48,64,77]. Even though *Flavonifractor plautii* was recently found to suppress the immune response in mice in multiple studies conducted by the same research group [101–103], it was repeatedly associated with several diseases, including ulcerative colitis, autoimmune diseases, obesity, and even with a poor diet [104,105]. Our findings of elevated *Flavonifractor*, with no study finding opposite results, strongly support its negative influence on the host's health, including affective disorder.

#### *3.5. Proteobacteria*

Within the phylum Proteobacteria, most differences between depressed individuals and HC were found within order Burkholderiales of class Betaproteobacteria. Five studies reported a reduction of family Sutterellaceae, OTUs within this family, or genus *Sutterella* in depressed subjects [51,57,60,63,67]. Additionally, Peter et al. [55] reported an association between order Burkholderiales abundance and perceived stress, which is inconsistent with the other results on lower taxonomic levels, demonstrating that a high taxonomic resolution should be striven for. No study found elevated Sutterellaceae or *Sutterella* in patients with affective disorder. Hence, a negative association with depression is conceivable on these taxonomic levels. Even though not very much is known about this family, it seems to be associated with diseases like autism spectrum disorder, down syndrome, and IBD. Furthermore, a mild pro-inflammatory capacity of certain species within genus *Sutterella* was proposed [106]. Therefore, the origin and consequence of reduced Sutterellaceae and *Sutterella* in depressed individuals remain unclear.

Within class Gammaproteobacteria, family Enterobacteriaceae tended to be elevated in subjects with affective disorder, but with controversial results. At the genus level, few studies reported a higher abundance of *Enterobacter* and *Klebsiella* to correlate with worse health [54,62,75]. The family Enterobacteriaceae, with its well-known genera *Enterobacter*, *Escherichia*, *Klebsiella*, *Salmonella*, and *Shigella*, is associated with many different clinical syndromes and diseases, including foodborne infectious diarrhea, enteritis, colitis, hemolytic uremic syndrome, as well as extraintestinal diseases [107]. Maes et al. found increased serum immune globulin M (IgM) against LPS of Gammaproteobacteria in depressed individuals, highlighting the link between intestinal mucosal dysfunction, increased bacterial translocation, immune response, and depression [9]. Family Pseudomonadaceae and genus *Pseudomonas* were elevated in depressed subjects in two studies, but with controversial results regarding only MDD subjects [46,58]. Maes et al. also found increased IgM against *Pseudomonas* in MDD subjects, compared to HC [33]. Here too, as concluded for other families, more detailed information on lower taxonomic levels is needed for a clear-cut statement.

Genus *Desulfovibrio* of class Deltaproteobacteria seemed to positively correlate with MDD and BD, but only three of all reviewed studies found a different abundance [44,57,74]. Cheng et al. analyzed published genome-wide association study data sets with high numbers of cases and controls [74]. They reported an association of genus *Desulfovibrio* with MDD, BD, and other mental disorders, suggesting a crucial role of this genus in mental

disorders. However, these findings are not consistent with the results of the other two studies, which found *Desulfovibrio* to be elevated only in female but not in male MDD subjects, and even reported an inverse correlation with MDD, respectively [44,57]. Age could be an important confounding factor, due to young participants in Liu et al. [44] and middle-aged in Chen et al. [57]. Therefore, genus *Desulfovibrio* could be reduced in young-aged, depressed subjects, while in middle-aged these bacteria could be elevated, but further investigation is needed.

#### *3.6. Actinobacteria*

Within the phylum Actinobacteria, study results tended towards an increase of class Coriobacteria, order Coriobacteriales, family Coriobacteriaceae, or OTUs within this family correlating with depression, but with inconsistent results [47,49,52,57,63]. However, on the genus level, a higher abundance of genus *Collinsella* was associated with lower anhedonia, BD treatment, and remission [47,48,66]. Only one study reported an association of elevated *Collinsella* with depression scores, suggesting that there is little evidence for a positive correlation with depression [57]. Other research found a stress-induced increase of an unspecified genus of Coriobacteriaceae in mice, a reduction of genus *Collinsella* after weight loss in obese type 2 diabetics, and a positive correlation of *Collinsella* with circulating insulin levels and low dietary fiber intake, while a high fiber intake supports SCFA-promoting gut bacteria [108–110]. Therefore, *Collinsella* is generally associated with worse health, and consequently, it remains unclear why especially genus *Collinsella* tended to be associated with ameliorated depressive symptoms. Even though closely related, genus *Eggerthella* was shown to be associated with MDD and higher depression and perceived stress scores in the reviewed research [43,51,57]. An elevated *Eggerthella* abundance was also found in immune-mediated inflammatory diseases like Crohn's disease and ulcerative colitis [111]. In conclusion, while on a higher taxonomic level an increase of these bacteria was found in depressed subjects, on lower taxonomic levels this consistent increase could not be seen, due to a reduction of genus *Collinsella* correlating with depression.

Within the order Bifidobacteriales, genus *Bifidobacterium* is known for its beneficial effects on the host's health and a lower abundance is associated with several diseases [112]. Counterintuitively, most studies found a positive association with depression and negative mood, while only three studies reported an elevation correlating with better health or depression treatment (Table 6). While in depressed subjects *Bifidobacterium* abundance seemed to be elevated, most studies reported a significant improvement in depressive symptoms with probiotics containing *Bifidobacterium* spp. (Table 7). The reason for those seemingly contradicting results remains unclear and needs further research.


**Table 6.** Different abundances of family Bifidobacteriaceae and genus Bifidobacterium.

Results tended to show a negative effect of an elevation of these bacteria. "↓" and green symbolizes a significant reduction in diseased subjects or inverse correlation with more severe symptoms, while "↑" and orange shows a significant elevation or positive correlation with more severe symptoms, and "=" and grey symbolizes no correlation. Brackets include additional information about the reported correlation (which bacteria showed a correlation or in which subgroup of patients a correlation was found). Empty cells symbolize that no significant results were reported. Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; HDRS, Hamilton depression rating scale; f, a correlation only in female subjects.

In general, most of the investigated bacteria belonging to phylum Actinobacteria tended to correlate with worse health and depression, which is again contrary to the general finding of Actinobacteria having a positive influence on human health and its beneficial effects as probiotics [112]. Microbial bacteria are firmly influenced by diet [112], but only few studies included dietary data. Therefore, different eating habits could be a possible factor leading to elevated Actinobacteria in depressed individuals, but a satisfactory explanation is not possible to date.

#### *3.7. Human Interventional Trials in Depression*

A total 13 studies investigated the influence of probiotic *Lactobacillus* and/or *Bifidobacterium* on depression. While six studies found no significant improvement in depressive scores, seven reported a significant amelioration of depression (Table 7). Three of them were conducted by the same research group, focusing specifically on the species *Lactobacillus gasseri* [114–116]. These three studies reported the most positive and most diverse results. According to them, *L. gasseri* ameliorated depression and anxiety, shortened sleep latency and awake time, lightened fatigue, improved global sleep quality, but also lowered salivary cortisol levels, and even suppressed unfavorable intestinal bacteria. However, to our knowledge, no other studies investigated the effect of *L. gasseri* supplementation on depressive symptoms to date. It might be essential to verify these highly encouraging results with *L. gasseri* probiotics by additional independent research groups. Supporting these beneficial findings of *Lactobacillus* and *Bifidobacterium* probiotics, Heym et al. reported a strong correlation between *Lactobacillus* spp. abundance and positive self-judgment, but only an indirect relationship between *Lactobacillus* spp. and depression, while *Bifidobacterium* spp. showed no association with any psychometric measures [80]. However, all but one of the compared studies reported a beneficial effect of these probiotics, and despite often not reaching significance level, depression scores mostly showed a slight reduction. This indicates that probiotic *Lactobacillus* and *Bifidobacterium* have a modest beneficial effect on depressive symptoms. Whether the effect size is large enough to be of clinical importance needs further investigation, but with none of these 13 studies reporting a worsening of depression or other serious side effects in the probiotic groups, probiotic *Lactobacillus* and *Bifidobacterium* should be considered as an adjunctive treatment in the therapy of affective disorder and depressive symptoms.

Additional studies investigated the use of prebiotics, synbiotics, and different probiotics on depression (Table 7). While in the probiotic group depressive scores only tended to ameliorate, in the synbiotic group a significant difference was found by Haghighat et al. [117]. Therefore, an additional supplementation containing fructo-oligosaccharides, galactooligosaccharides, and inulin could further support the beneficial effects of probiotic *Lactobacillus* and *Bifidobacterium*. In treatment-resistant MDD subjects, probiotic *Clostridium butyricum MIYAIRI 588*, a butyrate, acetate, and propionate producing bacteria, showed in combination with antidepressant medication not only an improvement of depressive symptoms but also a response rate as high as 70%, with 35% reaching complete remission [84]. These results are very promising and as mentioned before, further studies may pave the way for the use of probiotic *C. butyricum MIYAIRI 588* in depressive patients. The use of no probiotic food or supplementation was associated with higher odds of depression in a large cross-sectional study. However, individuals consuming probiotics were wealthier and showed a healthier lifestyle on average, resulting in a lower risk of developing depression [118]. These data indicate only an indirect link of probiotics and depression and emphasize that not all probiotic bacteria result in lower rates of depression.

#### *3.8. Studies Involving Twins and Their Relatives*

As genetics and environmental factors have a huge influence not only on the development of depression but also on the microbiome, three studies examined whether there is a correlation between the intestinal microbiota and depression in twins and relatives. Two studies investigated the difference of the microbiota in twins [53,54], and one study

examined the difference between the microbiota of patients with newly diagnosed BD and their first-degree relatives [64].


**Table 7.** Effect of prebiotics, probiotics, and synbiotics on depression.

Results mostly showed positive effects, but several studies could not find significant differences. "+" symbolizes a significantly positive health effect, while "=" indicates no significant difference. In brackets a selection of the investigated measures is given. Abbreviations: MDD, major depressive disorder; BD, bipolar disorder; D, participants with depressive symptoms; H, healthy participants; C, cross-sectional study; IBS, irritable bowel syndrome; BDI, Beck depression inventory; BAI, Beck anxiety inventory; DASS, depression anxiety stress scales; HADS, hospital anxiety and depression scale; PHQ-9, patient health questionnaire; HDRS, Hamilton depression rating scale; PSS, perceived stress scale; MADRS, Montgomery-Åsberg depression rating scale; BDNF, brainderived neurotrophic factor; QoL, quality of life; hs-CPR, high-sensitivity C-reactive protein; YMRS, young mania rating scale.

The two twin studies investigated 128 monozygotic twins and one pair of monozygotic twins, respectively [53,54]. Vinberg et al. distinguished between affected twins (with a diagnosis of MDD or BD in remission), unaffected high-risk twins (with a co-twin history of depression), and low-risk twins (without any histories of depression in the family) [53]. They found a lower diversity and richness of the microbiota of affected twins, while high-risk twins showed the same pattern, but with the lower diversity only being a trend. Affected and high-risk twins also showed an absence of an OTU belonging to the family Christensenellaceae. However, no correlation of the microbiota with illness severity was found. Jiang et al. reported a less similar microbiota between the pair of discordant twins (one twin with a history of depression and one without) than the microbiota of two healthy spouses [54]. Moreover, the similarity of the microbiota reached its maximum after achieving full remission of the affected twin, with the level of Ruminococcaceae and *Faecalibacterium* increasing and *Enterobacter* decreasing during the responsive and remission periods. Several SCFA-producing genera, mainly belonging to the families Lachnospiraceae and Ruminococcaceae, were reduced in the active-BD state compared to the healthy spouses [54]. Further, an over-representation of LPS biosynthesis genes in the gut microbiota during the active depressive period was found, whereas in the remissive state these genes decreased, indicating a potential recovery of the microbiota during the responsive and remissive period. But with only one pair of monozygotic twins, these results must be taken with caution.

According to Coello et al., newly diagnosed BD subjects had a different microbiota, while the microbiota of unaffected first-degree relatives did not differ significantly from the microbiota of HC [64]. Especially the presence of *Flavonifractor* was associated with an

increased odds ratio for having BD. After adjusting for smoking, this association attenuated, indicating an additional correlation of *Flavonifractor* with smoking, as well as with the female gender. It is hypothesized that the microbiota of BD patients is characterized by the different presence or absence of bacteria, especially of *Flavonifractor*, rather than the difference of bacterial abundance.

In summary, these studies support the hypothesis of a close connection between the intestinal microbiota and depression, and significant differences in the microbiota composition of depressed subjects were found. They even showed that a higher risk of developing depression is already associated with minor changes of the microbiota, and that with remission of depression, the intestinal bacteria change back towards a more "normal" composition. Comparing the reported bacterial alterations with our general findings, a beneficial effect of Christensenellaceae, Ruminococcaceae, and *Faecalibacterium*, and a negative effect of *Flavonifractor* show most experimental evidence.

#### **4. Discussion**

Our data show that the intestinal microbiota is closely linked with major depressive and bipolar disorder. The complexity of the microbiota makes it challenging to find clear causative associations, and a higher taxonomic resolution for determining the intestinal bacteria would be of importance for a more accurate analysis. Further studies with more participants are needed to verify specific bacterial alterations since the reviewed studies were mostly conducted with small sample sizes of up to 150 participants (with few exceptions). Study designs, inclusion criteria, analysis methods, and confounding factors varied widely, making a comparison difficult and may explain the contradicting results for certain bacteria. Therefore, a review of such heterogeneous studies is also associated with major limitations, and more standardized studies would facilitate a comparison.

Despite these limitations, this review demonstrated that certain bacteria consistently correlate with depression. The strongest and most consistent correlations are demonstrated in Table 8.


**Table 8.** Bacteria (with taxonomic level) that correlated most with depression.

Additionally, phylum Bacteroidetes consistently positively correlated with depression in young individuals, whereas in middle-aged individuals a strong inverse correlation with depression was found, while phylum Firmicutes showed opposite correlations.

Noticeably, apart from phylum Bacteroidetes, Firmicutes, and Actinobacteria, the strongest correlations with depression were found on low taxonomic levels (particularly on genus level), underlining the importance of high taxonomic resolution to identify bacterial alterations in depressed subjects. Further studies specifically focusing on these altered bacteria and their interactions with the host could provide a better insight into the connection between depression and the human gut microbiota. SCFA-producing bacteria were mostly found to be reduced in depressed individuals, emphasizing the beneficial influence on their host. With a better understanding of the intestinal microbiota, new

therapeutic strategies for the treatment of affective disorder could be found, which is crucial considering the high therapy resistance and relapse rates [5]. However, considering the complexity of the intestinal microbiota and the diversity of the bacterial changes found in this review, it is conceivable that bacterial clusters would show better correlations with depression. Consequently, studies with more participants are needed to identify depression-like bacterial clusters, as well as novel potential treatment approaches.

Changing the intestinal microbiota (for instance through specific diets, supplementations, probiotics, synbiotics, or fecal microbiota transplantation (FMT)) could potentially support the host's health and mitigate depressive symptoms. While multiple clinical studies found probiotics and synbiotics to have a positive impact on mood and behavior, clinical FMT studies in depressed subjects remain scarce. FMT has repeatedly been shown to ameliorate depression [127]. After receiving fecal transplants of depressive patients, a depression-like behavior of germ-free mice was observed, compared to mice receiving fecal transplants of healthy individuals [63,128]. In a clinical study, patients with gastrointestinal complaints reported an improvement in depression scores after FMT, and in a case report, a treatment-resistant BD patient achieved full remission after FMT [56,129]. Further studies exploring the effect of probiotics, synbiotics and FMT with more individuals are required to strengthen these positive findings. In addition, since diet significantly influences the intestinal microtioba composition [112] and only few studies included dietary questionnaires, it is essential to adjust for dietary changes between depressed subjects and HC. Thereby bacterial alterations only due to different eating habits could be excluded in future studies.

Stevens et al. were able to differentiate depressed and non-depressed subjects using a machine learning approach [130]. Additional similar studies could offer the potential of finding specific bacterial clusters and changes in metabolic pathways associated with affective disorder. Moreover, it is suggested by these authors that this novel approach may be used as a reliable diagnostic tool to identify different depression phenotypes in the future, potentially even leading to personalized treatment of depression [130]. Even though reliably distinguishing between depressed and non-depressed individuals, it remains doubtful whether similar approaches would be of clinical importance as a diagnostic tool.

Another key question is how depressed subjects develop different abundances of certain intestinal bacteria. Does it depend on the diet, with depressed individuals showing different eating habits, do they have a special intestinal milieu that secondarily favors the colonization of certain bacteria, or are there other factors influencing the intestinal microbiota towards a depressive-like composition? Answering this question, which is the scope of another dedicated report, would help to prevent unfavorable microbiota changes and would provide further information about the bidirectional connection of the microbiota and depression.

While most of the studies only investigated MDD subjects, research with BD subjects is lagging, but an increasing number of studies including BD individuals in recent years shows a growing interest. In this review, we could not identify unequivocal differences between the microbiota abundance of MDD and BD subjects. Three studies juxtaposed MDD and BD individuals and found a distinct microbiota, but the results were controversial and inconsistent with the other studies including only MDD or BD subjects [46,50,74]. Therefore, a distinguishable microbiota is conceivable, but major differences could not be found.

In conclusion, strong correlations between the intestinal microbiota and affective disorder were found. Specifically investigating only MDD or BD individuals would decrease the heterogeneity of the disease manifestation, but other factors such as the analysis methods, subject heterogeneity, medication, nutrition, and lifestyle factors essentially confound the results. Additional standardized research is needed to elucidate the connection between the intestinal microbiota and depression and to further examine their interdependencies to eventually find novel therapeutic approaches and lower the rates of treatment-resistant affective disorder.

**Author Contributions:** Conceptualization, T.K. and M.H.M.; methodology, T.K. and M.H.M.; validation, M.H.M.; formal analysis, T.K. and M.H.M.; investigation, T.K.; resources, T.K.; data curation, T.K.; writing—original draft preparation, T.K.; writing—review and editing, T.K. and M.H.M.; visualization, T.K.; supervision, M.H.M.; project administration, M.H.M. and T.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We would like to thank David Wolfer for useful comments and the review of this paper.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


#### **References**


**Shirley Mei-Sin Tran and M. Hasan Mohajeri \***

Department of Medicine, Institute of Anatomy, University of Zurich, Winterthurerstrasse 190, 8057 Zürich, Switzerland; tran.shirleymeisin@gmail.com

**\*** Correspondence: mhasan.mohajeri@uzh.ch; Tel.: +41-79-938-1203

**Abstract:** In the last decade, emerging evidence has reported correlations between the gut microbiome and human health and disease, including those affecting the brain. We performed a systematic assessment of the available literature focusing on gut bacterial metabolites and their associations with diseases of the central nervous system (CNS). The bacterial metabolites short-chain fatty acids (SCFAs) as well as non-SCFAs like amino acid metabolites (AAMs) and bacterial amyloids are described in particular. We found significantly altered SCFA levels in patients with autism spectrum disorder (ASD), affective disorders, multiple sclerosis (MS) and Parkinson's disease (PD). Non-SCFAs yielded less significantly distinct changes in faecal levels of patients and healthy controls, with the majority of findings were derived from urinary and blood samples. Preclinical studies have implicated different bacterial metabolites with potentially beneficial as well as detrimental mechanisms in brain diseases. Examples include immunomodulation and changes in catecholamine production by histone deacetylase inhibition, anti-inflammatory effects through activity on the aryl hydrocarbon receptor and involvement in protein misfolding. Overall, our findings highlight the existence of altered bacterial metabolites in patients across various brain diseases, as well as potential neuroactive effects by which gut-derived SCFAs, p-cresol, indole derivatives and bacterial amyloids could impact disease development and progression. The findings summarized in this review could lead to further insights into the gut–brain–axis and thus into potential diagnostic, therapeutic or preventive strategies in brain diseases.

**Keywords:** gut–brain–axis; gut microbiome; short-chain fatty acids; bacterial metabolites; SCFA

#### **1. Introduction**

We are exposed to bacterial organisms from the beginning of our existence to the end of it. Even before birth, bacteria have been detected in the meconium of newborns, thus discrediting the pre-existing idea of a sterile foetal stage [1]. Later on, the early postnatal exposure to either the mother's vaginal flora or microbes from the environment, depending on delivery, impacts microbial colonization patterns, overall health and the neurodevelopment of the individual [2]. Although the microbial residents in our gastrointestinal tract (GIT) have already been known to impact the state of human health, the theory of a bidirectional gut–brain–axis (GBA) has taken the spotlight of global researchers mostly after the turn of the millennium.

Individuals are globally affected by increasing morbidity and mortality of psychiatric, neurodegenerative and neurodevelopmental disorders. The aetiology and pathophysiology of these brain diseases remain to this day to be fully elucidated and treatment options are largely of symptomatic nature. Therefore, researchers have unsurprisingly been looking at novel perspectives of disease, such as the GBA. Emerging findings on gut microbial influence on our nervous system were reported, involving bacterial-derived toxins, vitamins and neurotransmitters, yet the precise mechanisms, the "language of the GBA" [3], remain to be fully elucidated. Some newly examined neuroactive bacterial metabolites have nevertheless shown potential to play a role in this communication (Figure 1).

**Citation:** Tran, S.M.-S.; Mohajeri, M.H. The Role of Gut Bacterial Metabolites in Brain Development, Aging and Disease. *Nutrients* **2021**, *13*, 732. https://doi.org/10.3390/ nu13030732

Academic Editor: Franck Gael Carbonero

Received: 19 January 2021 Accepted: 22 February 2021 Published: 25 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Figure 1.** Non-exhaustive overview of neuroactive gut bacterial metabolites. SCFAs = short-chain fatty acids; BA = butyric acid; PA = propionic acid; AA = acetic acid; VA = valeric acid; iBA = isobutyric acid; iVA = isovaleric acid; iCA = isocaproic acid; TMAO = trimethylamine N-oxide; 3-HBA = 3 hydroxybenzoic acid; 3,4-diHBA = 3,4-dihydroxybenzoic acid; DHCA = dihydrocaffeic acid; IS = indoxyl sulphate; 4EPS = 4-ethylphenylsulfate.

about the GBA's role in brai This systematic review intends to summarize the research on various families of neuroactive bacterial metabolites as probable key players in the GBA. The focus is their effects on disorders of the brain, ranging from neurodevelopmental stages in childhood to neurodegenerative diseases in advanced age. Although intriguing evidence has emerged about the GBA's role in brain tumorigenesis via the modulation of the immune system, we refer the reader to a recent extensive study [4], as a detailed examination of this subject is beyond the scope of this review. Considering the magnitude of various influences from bacterial metabolites on the human organism, we will focus mostly on direct neuroactive effects on the brain. Most papers have largely emphasized taxonomic shifts in gut microbiota in specific diseases, or short-chain fatty acids (SCFAs) to date. One of our objectives is to provide a summary of findings between SCFAs and brain diseases, while in the second part of this review, reports of less explored non-SCFAs will take centre stage.

#### **2. Materials and Methods**

This systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [5]. The main objective was to explore and summarize the available data on influences of gut bacterial metabolites on the brain, with a focus on neurodevelopmental, autoimmune-mediated neuroinflammatory, and neurodegenerative diseases.

terms "bacterial metabolites" combined with "brain development", "brain aging", "brain ageing", "brain disorders", "brain diseases", "neurodegenerative", "neuroprotective", "gut brain axis" and "gut axis" delivered 216 hits after The first PubMed and SCOPUS databank searches were conducted on 20 November 2019. A second search was performed on the 8 July 2020 with the objective to include additional recently published data. The following search parameters and MeSH (Medical Subject Headings) terms "bacterial metabolites" combined with "brain development", "brain aging", "brain ageing", "brain disorders", "brain diseases", "neurodegenerative", "neuroprotective", "gut brain axis" and "gut-brain-axis" delivered 216 hits after removing duplicates (Figure 2). The second search with the same search parameters delivered 76 new hits. One hundred and forty-seven additional records with relevant information were individually selected from the list of references of the initially identified papers. Our focus on gut bacterial metabolites warranted the exclusion of data on viruses, archaea, and fungi

• • • •

as well as data on bacteria not related to the gut microbiome. Original papers as well as reviews were included, while no restriction on publication year was applied. The inclusion criteria were the following:


**Figure 2.** Methodical approach of our systematic review adhering to Preferred Reporting Items for Systematic Reviews **Figure 2.** Methodical approach of our systematic review adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria (PRISMA criteria [5]).

Most of the papers were dated from 2013 to 2019. Three papers lacking full texts, as well as two non-English publications, were excluded (Figure 2). Two hundred and seven papers were further excluded based on the lack of relevance to the topic. Finally, 227 studies were inspected for the qualitative synthesis. As to our knowledge, no other review to date has undertaken an analysis to this extent of links between several categories of gut bacterial metabolites and brain diseases.

#### **3. Short-Chain Fatty Acids**

– Short-chain fatty acids (SCFAs) are saturated fatty acids produced by the bacterial fermentation of dietary fibre [6]. The majority of SCFAs consist of acetic (AA), propionic (PA) and butyric acid (BA), which are mostly deprotonated in the intestine (acetate, propionate, butyrate) [7]. Some gut bacterial species capable of generating SCFAs are *Bacteroides*, *Bifidobacterium*, *Propionibacterium*, *Eubacterium*, *Lactobacillus*, *Clostridium*, *Roseburia* and *Prevotella.* Among them, *Roseburia*, *Eubacterium* and Lachnospiraceae (Firmicutes phylum, Clostridia class) are strong BA producers, while AA producers belong to the *Bifidobacteria spp.* [8]. Considering their production site, the initial point of contact with the human organism are colonocytes and other intestinal cells. This naturally leads to discussions of local impacts from SCFAs on overall gut health, predominantly in the context of diseases like irritable bowel syndrome (IBS) [9–11] and the inflammatory bowel diseases Morbus Crohn and Colitis Ulcerosa [12,13]. Local effects facilitated by SCFAs have previously been discussed in detail and will not be further elaborated in this review [12,14,15].

It is known that SCFAs are able to modulate gut permeability by upregulating tight junction proteins [16,17], which are also part of the blood–brain barrier (BBB). This conceivably raises the idea that barrier integrity of gut and brain could be similarly affected by

SCFAs [18]. Indeed, studies in germ-free (GF) mice demonstrated that SCFAs are capable of modulating BBB permeability, which consequently impacts the extent to which beneficial or harmful molecules in circulation can reach brain tissue [19,20]. For example, physiological amounts of PA have been recently shown to protect the BBB from oxidative stress [21] and to decrease paracellular permeability [22]. Similarly, BA and BA-producing *Clostridium butyricum* can lower BBB permeability through enhancing tight-junction expression in mice [22]. In addition to directly affecting the BBB, SCFAs might actually reduce systemic inflammation by decreasing gut permeability, thereby decreasing circulating gut-derived bacterial components that trigger neuroinflammation by injuring the BBB or by affecting immune cells and cytokines in the brain [15]. SCFAs also act upon various gut–brain– pathways including immune, endocrine, vagal and direct humoral pathways (extensively reviewed by Dalile et al. [15]) and some effects in cellular systems, namely:


These studies have demonstrated SCFAs to be capable of regulating neuroinflammatory processes involving immune cell recruitment and cytokine secretion [29]. Microglia, immune cells residing in the CNS, were observed to be dysregulated in various psychiatric disorders like depression, schizophrenia, autism spectrum disorder (ASD) and obsessive– compulsive disorder [15] as well as in germ-free animals [28]. Interestingly, Toll-like receptors (TLRs) known to recognize bacterial compounds and to regulate inflammatory responses in our gastrointestinal tract (GIT) were found on various cell-types of the CNS, thus further supporting a link between gut and brain immune processes [30]. SCFAs also seem to directly impact neuronal function, as reported by studies showing PA and BA affecting intracellular potassium concentrations [31] and findings on influences on neurotransmitter metabolism [15]. Furthermore, beneficial effects on preserving memory function in experimental meningitis and protection from ouabain-induced hyperlocomotion were reported through a modulatory effect on the expression and activity of neurotrophic factors like brain-derived neurotrophic factor (BDNF), nerve growth factor (NGF) and glial cell line-derived neurotrophic factor (GDNF) in rats [32,33]. Interesting to note are the effects on synaptic plasticity by HDACI, since this process involves protein synthesis and therefore, gut-derived SCFAs might be potential epigenetic modulators of learning, memory formation and storage [34,35]. In light of these findings, altered SCFA production in the

presence of gut microbiome disturbances, also known as dysbiosis, has been postulated as a potential risk for brain developmental and neurodegenerative diseases. Currently, CNS pathologies are often associated with changes in taxonomical gut microbiome and bacterial metabolites, as will be elaborated on in the following chapters.

#### *3.1. SCFA and Autism Spectrum Disorder*

Autism spectrum disorder (ASD), a neurodevelopmental disorder characterized by behavioural abnormalities including repetitive behaviour, communication deficits and sensitivity to environmental changes, is often linked to gastrointestinal problems and alterations in the gut microbial community [36–39]. This was shown in a cohort of human infants, that distinct gut bacterial composition variations, at times called enterotypes, might correlate with cognitive performance [40]. Recent studies indeed reported the gut microbiome compositions of children with ASD to be significantly distinct from their neurotypical (NT) developing peers, and furthermore detected overall lower alpha diversity in ASD gut microbiomes [41].

Only a handful of human studies have measured faecal metabolites in ASD, with some of them reporting elevated [42] and others decreased total SCFA levels [39,43] in children with ASD (Table 1). Contrarily, Kang et al. [41] reported no significant differences in SCFAs levels between ASD and NT control group. Adams et al. [43] reported lower faecal levels across all SCFAs (AA, BA, PA, valeric acid (VA)) in children with ASD. Others observed lower levels of AA and BA, but no significant alterations in PA-levels [44]. Conversely, significantly increased faecal levels across all SCFAs (AA, BA, PA, VA, isobutyric and isovaleric acid) in one study [42], and significantly elevated AA and PA in another study were detected in ASD faecal samples [39,43]. In support of the findings on decreased BA in two of the studies mentioned, a metagenomic analysis on faecal samples resulted in a lower abundance of microbial genes involved in the production of BA [38], which also parallels a prior reported decrease in BA-producing *Faecalium prausnitzii* in autistic patients [41].

Supporting the idea of gut microbial influence on ASD development, Sharon et al. [45] demonstrated that faecal microbiome transplants (FMT) from human ASD donors were able to invoke ASD-like behavioural traits in mice. Moreover, El-Ansary et al. [46] reported neuronal DNA damage induced by PA oral administration in a hamster model. These suggested that PA could play a role in neurotoxicity by damaging mitochondrial DNA by ATP-depletion, thus leading to mitochondrial dysfunction and oxidative stress in neurons. This postulated pathway in autism has been underscored by earlier findings in rat pups exposed to PA, exhibiting various immune, mitochondrial and ASD-like behaviour changes similar to ASD in humans [46–49]. PA-induced ASD in rodents is a validated model for ASD research that has presented with abnormal neural cell organization and hippocampal histology, increased microglia activity, neurotoxic cytokine secretion, and typical ASD-like behaviour traits [50]. Moreover, perturbed microbiota with increased PA-producers and decreased BA producers correlated with the severity of disease burden in ASD [26,43], even if studies of PA faecal levels in children with ASD compared to healthy controls (HCs) have produced conflicting data [39,41].

Contrarily to PA, BA has shown overall beneficial effects in ASD. BA administration alleviated ASD-like behaviour and normalized changes in gene transcription related to inhibitory/excitatory balance in the frontal cortex of the T+tf/J strain of the black and tan brachyury (BTBR) mouse autism model [51]. Nankova et al. [23] reported that SCFA as epigenetic regulators might affect genes assumed to be involved in ASD. BA and PA were able to increase catecholamine production as HDACI by regulating the tyrosine hydroxylase (TH) gene in an in vitro neuronal cell line (PC12 cells). PA and BA also modulated lipid homeostasis and inflammatory processes [23]. Moreover, SCFAs' influence on various genes of the dopaminergic pathway were detected, specifically on dopamine beta-hydroxylase (DBH) which, when dysregulated, shows associations with ASD in humans [23]. Interestingly, the serotonin system has been shown to only be affected by the administration of PA [52]. Furthermore, the study presented downregulating effects by PA or BA in the expression of fragile X mental retardation 1 (FMR1), neurexin and neuroligin, genes previously reported to relate to ASD [53–56]. BA, among all SCFAs, is the most important HDACI to modulate brain function through epigenetic processes [57] and thus, altered BA levels might potentially modify neuronal function.

In addition to the potential role of BA and PA in ASD, it is worth noting that the structurally related valproic acid (VPA), a branched SCFA, effectively creates a frequently used ASD mouse model that mimics both behavioural as well as gut microbiome traits in ASD patients [58]. In addition, prenatal exposure to VPA significantly increases the risk of ASD and showed epigenetic effects on neurotransmitter homeostasis via HDACI, similarly to BA and PA [59–62]. Additionally, VPA invokes dysfunctions in glutamate and GABA-neurotransmission and is thus likely to produce an altered balance between excitation and inhibition in the cerebral cortex [63].

These studies have shown that alterations of SCFAs can intricately influence neurodevelopmental processes via epigenetic modulation as HDACI. In support of a connection to ASD, Stilling et al. [64] detected upregulated cAMP response element-binding protein (CREB)-dependent gene expression in amygdala of GF mice, a limbic structure involved in emotion, memory and behaviour. It is thus understandable that a dysfunctional amygdala has been associated with neuropsychiatric disorders like anxiety disorder, post-traumatic stress disorder (PTSD) and ASD [65].

In contradiction with the above data, no significant changes in SCFA production were found in GF mice inoculated with microbiota from poor growth and good growth preterm infants, even though the administered microbiota was associated with pathologic developmental changes in neurons and oligodendrocytes of the receiving mice [66]. This might point to a different and/or additional pathway than SCFA, by which gut microbiota may affect early neurodevelopment.

Overall, support for SCFAs as putative influencers on ASD are present in a handful of clinical and mainly preclinical studies, though the research is still in its infancy. Therefore, further investigation to bring light into this emerging theory is strongly recommended.

#### *3.2. SCFAs and Affective Disorders*

Pathophysiological factors in affective disorders are multifaceted and gut microbial involvement has gradually become a potential contributing factor. Faecal SCFA levels from humans [67,68] and primates [67,68] with major depressive disorder (MDD) showed an overall decrease and altered composition compared to HCs (Table 1). AA, PA and isovaleric acid significantly decreased while only isocaproic acid increased in faecal samples of depressed individuals [67,68]. In contrast, one study reported no significant changes in faecal SCFAs in depressed patients [69]. Nevertheless, researchers previously showed distinct differences between faecal microbial compositions of HCs and MDD through taxonomic association studies [70]. Further links between affective disorders and a disturbed gut environment might be provided through observations in functional gut disorders like irritable bowel syndrome (IBS), exemplified by the results of a recent meta-analysis with significantly increased anxiety and depression in IBS patients [71].

A mentionable study by Kelly et al. [69] presented that depressive behaviour can be transferred from humans to germ-free rats by FMT, suggesting a strong connection between gut bacteria and affective disorders like major depressive disorder (MDD). Interestingly, there were discrepant findings regarding the role of SCFAs: faecal AA and total SCFA levels were higher in rats receiving FMT from patients than from HCs. However, depressed and healthy human donors showed no significant differences in their faecal SCFA levels. This calls for further investigations in clinical studies since interspecies differences might be a contributing factor in this case. Recently, rats bred for high anxiety-like behaviour (HAB), an animal model for anxiety and depression, displayed lower microglia numbers in distinct brain regions (infralimbic and prelimbic prefrontal cortex) and gut microbial shifts toward decreased counts of the BA-producing Lachnospiraceae family [72]. Treatment with antibiotic minocycline alleviated male HAB rats of depressive symptoms, further decreased

circulating inflammatory cytokines and microglial count, as well as enriched their microbiota with known BA and 3-OH-butyrate producers Lachnospiraceae and Clostridiales family XIII. In fact, Clostridia are considered as the main BA-producing class of the human gut microbiome [73] (Table 2). These findings, together with previous propositions for immunomodulatory effects of BA and 3-OH-butyrate on inflammation, T-cell and microglial activity [13,28,29,74,75] point towards an intricate relationship between microbial derived SCFAs and affective disorders, that might benefit from their anti-inflammatory effects. In support of this theory, increased markers of inflammation such as pro-inflammatory cytokines in circulation and the brain are correlated with MDD [76]. Moreover, studies have successfully demonstrated SCFA-mediated anxiolytic and antidepressant effects in mice undergoing induced psychosocial stress [77]. In particular, the administration of sodium butyrate (NaB, the sodium salt of BA) has been reported to alleviate pathologic affective behaviours in rat models, including hyperactivity, depressive and manic symptoms [26]. Future work on this subject, especially through metabolomic studies in humans, might enlighten the intricate gut bacterial metabolite–brain axis interplay in affective disorders, as the current state of research provides only few clinical studies on this particular subject.

#### *3.3. SCFAs and Autoimmune Diseases of the Brain: Multiple Sclerosis (MS)*

Multiple sclerosis (MS) is an autoimmune disease of the CNS that mainly damages the myelin sheaths of motor neurons. An imbalance between anti-inflammatory Treg cells and proinflammatory Th1 and Th17 cells are widely understood to take part in the MS pathophysiology [78].

Individuals with MS have been reported to harbour microbiomes that are significantly different from HCs [79,80]. Indeed, one recent study reported increased *Streptococcus,* decreased *Prevotella\_9* and overall decreased faecal SCFAs (AA, PA and BA) in a Chinese cohort of MS patients [81]. *Streptococcus* is known to produce all SCFAs [44,82] and *Prevotella\_9* is able to generate AA and PA [81] (Table 2). MS patients displayed higher abundance of inflammatory Th17 cells, as anti-inflammatory Treg cells were decreased. Interestingly, faecal SCFA concentrations positively correlated with levels of circulating Treg cells in this study, thus suggesting that SCFAs exert anti-inflammatory effects due to elevated Treg/Th17–cell ratios. Similarly, significantly decreased SCFAs—were detected in blood samples of patients with active secondary progressive MS [29]. These two human studies might suggest an overall decrease in faecal and consecutively depleted circulating SCFA levels in MS patients (Table 1), that might shift the immune system towards proinflammatory processes due to lower Treg/Th17 cell ratios.

Autoimmune processes in the CNS were affected by the gut through SCFAs and longchain fatty acids (LCFAs) in the experimental autoimmune encephalomyelitis (EAE) mouse model of MS. The differentiation of pro-inflammatory Th1 and Th17 cells were increased by LCFAs, while anti-inflammatory Treg cell differentiation was boosted by SCFAs through the downregulation of the JNK1 and p38 pathway. Therefore, LCFAs exacerbated, while SCFAs alleviated disease and subdued axonal damage. Additionally, PA demonstrated the most stimulating effect on Treg cell differentiation, which improved histopathological outcomes of the spinal cord in EAE mice [13]. Melbye et al. [83] reviewed two other studies in EAE mice, who supported the ameliorating role of SCFAs in disease activity by modulating an increase in anti-inflammatory Treg cells and a decrease in pro-inflammatory Th1 and Th17 cells. BA too, was able to ameliorate demyelination in rats and importantly, exposing an organotypic slice culture to BA resulted in suppressed lysolecithin-induced demyelination and enhanced remyelination, represented by higher counts of mature oligodendrocytes [84]. In congruence with these studies, a recent review concluded that PA and BA ameliorated the clinical symptoms of EAE by inducing immune tolerance epigenetically as HDACIs. The proposed mechanism involves an upregulation of the transcription factor Foxp3 leading to increased Foxp3+ T regulatory lymphocytes, also known as Treg cells that inhibit proinflammatory Th1 and Th17 cells [85]. In addition to these findings that support an overall anti-inflammatory effect through SCFAs, Park et al. [29] recently demonstrated

that SCFA administration to EAE mice models increased anti-inflammatory IL10+Tcells and IL-10, as well as pro-inflammatory Th1, Th17 and Tc cells. Moreover, SCFA receptors GPR41 and GPR43 have demonstrated proinflammatory effects in EAE pathogenesis [29]. These results underline the importance of SCFAs to protect from inflammatory processes in the CNS. Their uncovered pro-inflammatory effects, however, indicate a complex system in immunomodulation, which calls for further work in this subject in order to evaluate potential interventions involving SCFAs in neuroinflammatory diseases.

#### *3.4. SCFAs and Neurodegenerative Diseases of the Brain*

Neurodegenerative diseases are becoming increasingly prevalent as the population gradually grows older. Researchers are trying to elucidate the pathomechanisms of the various brain diseases including Alzheimer's disease (AD), Parkinson's disease (PD), dementia with Lewy bodies (DLB), multiple system atrophy (MSA) and Huntington's disease (HD) [86]. This chapter will first briefly list some findings on SCFAs and neurodegenerative processes in general before focusing on AD and PD.

#### 3.4.1. General Findings on Neurodegenerative Processes

A recent in vitro study investigated the direct influences of the SCFAs NaB, sodium valerate and hexanoic acid on neuroinflammation and found that high concentrations of NaB were able to decrease the basal levels of the proinflammatory cytokine IL-6 in human glioblastoma–astrocytoma U373 cells [87]. However, further findings showed no neuroprotection from induced oxidative stress in differentiated SH-SY5Y cells (humanderived neuroblastoma cells) by any SCFAs. Interestingly, exposure to BA and valerate was able to induce neuronal maturation through MAP2-gene expression in undifferentiated neuroblastoma cells, thus hinting towards a beneficial effect on neurogenesis [87]. BA's effects in animal models include the potential to alleviate impaired cognition, enhancing neuronal plasticity, improve learning and memory performance, as well as neuroprotection, all beneficial processes regarding neurodegenerative diseases [57].

Overall, direct impacts on brain cells by SCFAs seem to be complex as well as dosedependent, which supports a hypothesis that anti-inflammatory processes in the brain, neuroplasticity and neurogenesis could be positively modulated through the manipulation of gut bacterial production and/or external supplementation of SCFAs. Recent research further provided evidence for an ameliorating role of SCFAs in inflammatory hippocampal neurodegeneration in mice through the reduced impairment of the intestinal barrier, which was induced by a high-fructose diet. It was suggested that SCFAs could amend the faulty colonic NLRP6 inflammasome responsible for epithelial impairment to alleviate hippocampal neuroinflammation, thus possibly reducing the likelihood of neurodegenerative processes associated with a typically high-fructose Western-style diet [88]. This might be an indirect mechanism by which SCFAs could exert neuroprotective effects.

#### 3.4.2. SCFAs and Alzheimer's Disease

Gut microbiome of Alzheimer's disease (AD) patients were observed to be altered, with decreased overall richness and diversity as well as some shifts within taxonomical compositions [89,90]. Some studies presented AD progression to associate with dysbiosis and that a healthy gut microbiome provides beneficial effects in AD patients and rodent models [90,91]. Recent studies further showed significant changes in gut microbiome compositions between AD patients and HCs at the genetic level, suggesting some bacterial AD-associated PCR products to be a potential marker of AD risk [92]. As Franceschi et al. described in their review in 2019, disturbances in the gut microbiome might influence processes involved in AD pathogenesis, such as chronic inflammation, molecular mimicry and Aβ accumulation. Furthermore, the presence of microbiome enterotype III (low *Bacteroides* and *Prevotella*) and the absence of enterotype I (>30% *Bacteroides*) were reported with stronger associations to the presence of dementia than classic markers (Table 3) [93]. This highlights the potential of the GBA to impact pathogenesis in dementia, though

unfortunately, no human studies that measured SCFA faecal levels have been reported as to our literature search.

The GF condition in transgenic AD mice models were observed to slow the progression of disease symptoms [94], underlining an important role for the presence of the gut microbiome, including bacteria and their metabolites in AD pathogenesis. A study with the APP/PS1 mouse model of AD reported disturbed microbiota composition and diversity, as well as overall lower SCFAs levels compared to wild-type (WT) controls. Additionally, over 30 metabolic pathways possibly related to amyloid deposition and ultrastructural anomalies were detected in intestine samples of the AD group [95]. Zheng et al. [96] have introduced a method of stable isotope labelling and liquid chromatography–tandem mass spectrometry to sensitively detect 21 SCFAs in mice faecal samples of AD and WT mice. In an AD mouse model, decreased levels of PA, isobutyric acid, 3-hydroxybutyric acid, and 3 hydroxyisovaleric acid were detected while increased levels of lactic acid, 2-hydroxybutyric acid, 2-hydroxyisobutyric acid, levulinic acid and valproic acid were found. In contrast to these findings, faecal PA was enriched in mice receiving FMT from an AD donor in comparison to a healthy one [97]. However, two faecal donor samples selected out of groups of 14 healthy and 13 AD volunteers might limit that study's evidential impact due to putative inter-individual variations.

The prevention of Aβ accumulation and the removal of accumulated amyloid plaque have been at the core of anti-AD therapeutic undertakings for more than two decades [98]. It is important to highlight an in vitro study reporting that valeric acid (VA), BA and PA, but not isobutyric acid, isovaleric acid and AA, to be capable of stopping the misfolding of Aβ40 peptides to neurotoxic Aβ40 aggregates in a dose-dependent manner [99]. Additionally, the same experiment on Aβ42 aggregation showed that only VA could inhibit the process dosedependently. A third experiment determined that VA and BA successfully halted Aβ fibril formation in a dose-dependent manner. These results demonstrate a mechanism by which gut microbial-derived SCFAs may benefit AD patients and that a gut microbiome depleted of SCFA producers might promote neurotoxic amyloid build up in the CNS. In support of this theory, Sun et al. [100] reported that FMT from WT-mice to the APP/PS1 mice model of AD resulted in the alleviated brain deposition of Aβ as well as levels of neurotoxic Aβ40 and Aβ42, tau protein phosphorylation, synaptic dysfunction, neuroinflammation and cognitive deficits, accompanied with restored alterations in gut microbiota and faecal SCFA levels. The AD mice harboured a perturbed microbiome enriched with Proteobacteria, Verrucomicrobio (phylum level), and *Akkermansia*, *Desulfovibrio* (genus level), with depleted Bacteroidetes phyla. All these conditions were reversed through FMT treatment. However, these microbial changes were lacking consistency in the relative abundance of bacterial species, for example a relative increase in Bacteroidetes or BA-producing Firmicutes has been previously observed in animal and human studies of AD [91]. Therefore, definite conclusions about distinct AD gut microbiome compositions and their capacity of SCFA production cannot be made at this point in time, which further warrants our focus on disease correlations with bacterial metabolites instead.

Impaired epigenetic gene expression has been discussed as a key factor in AD pathogenesis [101], which conceivably led to a study of BA's role as HDACI in an AD mouse model. Treatment with BA was able to improve associative memory function at an advanced stage of disease [102]. Other studies mentioned the neuroprotective capacity of BA to manipulate regulatory regions of the Forkhead box gene locus as HDACI. This provides a preventative and/or therapeutic potential to affect the balance between life-promoting and apoptotic cell processes critical in neurodegenerative diseases [8]. BA as NaB has shown neuroprotective benefits as HDACI in studies of PD, AD and HD, particularly leading to improved learning and memory in dementia, the prevention of oxidative stress and neuronal cell death in HD and PD, as well as overall upregulated transcription of neurotrophic factors involved in plasticity, survival and regeneration [103]. These results might indicate that decreased or overall altered gut microbial SCFAs and thus, dysregulated histone-acetylation, might indeed be connected to AD and related brain diseases. We

therefore suggest future studies to look for putative impacts of altered SCFA-producing gut microbiota on AD-related epigenetic processes in the brain. SCFAs might also impact AD indirectly through additional pathways via the regulation of intestinal gluconeogenesis by FFAR3 signalling, which affects the activity of the dorsal motor nucleus of the vagus, a structure with altered activity in PD and AD [8]. BA especially has also been hypothesized to positively impact cognition in AD patients via the stimulation of vagal afferents [8].

In a study with rats fed a high-fat diet, it was shown that the administration of two valeric acid esters (monovalerin and trivalerin) led to higher levels of AA in the brain, serum and liver, while caecal levels decreased. These data suggest that AA can actually be increased in the brain by oral supplementation and uptake in the gut [104]. This might be of interest, since AA administration to lipopolysaccharide (LPS)-stimulated astrocyte cultures was successful in producing anti-inflammatory effects [105]. BA exposure invoked anti-inflammatory effects as well, as shown by the reduced microglial activation and decreased secretion of inflammatory cytokines. BA inhibits the secretion of HDAC gut microbe-derived circulating inflammatory cytokines and thus limits their effects on neuroinflammatory processes that have been postulated to be involved in AD pathology [91]. Pro-inflammatory cytokines derived from dysbiosis might invoke the formation of Aβ aggregates as well as cause the dysfunctional maturation of microglia, thus leading to increased amyloid accumulation in the CNS [91]. Taken together, healthy gut flora with undisturbed SCFA production might benefit AD patients with decreased neuroinflammation and amyloid accumulation.

#### 3.4.3. SCFAs and Parkinson's Disease

PD is, after AD, the second-most prevalent neurodegenerative disease in the world [106] and is part of a cluster of neurodegenerative disorders associated with aggregated amyloid proteins. Misfolded alpha-synuclein proteins (αSyn) are specifically implicated in PD, DLB and MSA, also jointly known as "Synucleopathies" [107]. In PD, the dopaminergic neurons residing in the substantia nigra pars compacta are lost, subsequently leading to impaired motor functions [108]. Gut dysbiosis and GI dysfunction have been repeatedly mentioned as a hallmark of PD [108–114], thus investigations of mechanistic processes involving the GBA have emerged in recent years. This conceivably led to questions about gut microbial participation in pathophysiological processes of PD, such as the spreading of αSyn aggregates from gut to brain via the vagal nerve [115] as well as probable connections between gut dysbiosis, neuroinflammation and misfolding of αSyn [116].

Overall decreased SCFA levels with relatively low BA and a microbiome with reduced Bacteroidetes, Prevotellaceae as well as enriched Enterobacteriaceae were reported in PD [117] (Table 1). Underlining these findings, a recent review reported trends of reduction in SCFA producers in a PD patient's microbiomes, specifically reduced Lachnospiraceae (*Blautia*, *Dorea*, *Coprococcus*, *Rosburia*, *Clostridium XIV*), *Faecalibacterium* and *Bacteroides* [109]. Interesting to mention is the overall increased abundance of Enterobacteriacea, a phylum that is known to produce SCFAs (Table 2) and to also associate with the severity of motor symptoms in PD patients [112]. This finding might at the first glance appear counterintuitive under the assumption that SCFAs and their producers are beneficial to PD. On the other hand, the relative abundance in Enterobacteriaceae might further indicate the production of other metabolites involved in PD, as will be elaborated on later in the chapter discussing bacterial amyloids.

Two studies in rodents reported further contradicting results regarding SCFA levels in PD. Sampson et al. [118] used a transgenic αSyn-overexpressing mouse model of PD, that presented ameliorated PD pathologies when in a germ-free (GF) state or treated with antibiotics (AT). These GF/AT mice were then inoculated with human PD-donor microbiota. This treatment significantly altered faecal microbial communities and SCFA composition, displaying lower AA, but higher PA and BA, as well as worsened motor dysfunction compared to those receiving healthy FMT. Thus, the administration of a mixture of SCFAs to GF/AT mice was effective in inducing motor deficits, as well as

αSyn aggregation and microglial activation in the brain. This suggested a relevant role for SCFAs as mediators of PD in a genetically susceptible animal model [118]. Supporting these findings, the 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine(MPTP) – induced PD mice model presented an increased abundance of faecal SCFAs. The gut microbiome of this PD model was administrated to normal mice, which resulted in motor impairment and decreased striatal neurotransmitters, while FMT from healthy donors alleviated those symptoms [119]. The inconsistencies within the previously mentioned studies in human subjects, regarding beneficial or detrimental effects of SCFAs in PD, might point towards inter-species differences of mice and humans and the GF state of acutely inoculated mice. This could further suggest that even though the presence of SCFAs seems necessary to trigger pathological changes in genetically vulnerable organisms, shifts towards depleted SCFA levels and their bacterial producers might play a role in already established PD.

Several lines of evidence suggest that SCFAs, BA in particular, may exert possible beneficial effects in PD. First, BA might play a role in PD as a neuroprotective agent due to its agonistic effect on the receptor GPR109A, which promotes anti-inflammatory processes [120]. In addition, BA might also benefit PD patients with reduced neuroinflammation, indirectly enhanced dopamine synthesis through increased free niacin levels, as well as improved energy homeostasis and mitochondrial function [103,121]. Lastly, SCFAs ameliorated dysfunctional microglia in GF mice, which was represented by improved microglial maturation, morphology and function [28]. Proper mature microglial function includes decreased inflammatory activity and phagocytosis for amyloid proteins like tau, Aβ, and αSyn. Therefore, the state of the gut microbiome and its production power for SCFAs might positively influence several aspects of neurodegenerative diseases [91]. It might be of interest that mice lacking the SCFA receptor FFAR2 have shown dysfunctional microglia similar to GF animals, however, that particular study suggested alternative pathways by which SCFAs directly exert their effects on microglia due to a lack of evidence for FFAR2-expression on CNS cells [28]. Definite mechanisms involved in receptor-mediated processes of SCFAs remain to be determined.

As previously mentioned, SCFAs can upregulate gene-expression as HDACI. This process was shown to facilitate neuroplasticity and long-term memory, involving CREBdependent gene regulation [122,123]. In vitro studies also discovered PA and BA to modulate transcription of the tyrosine hydroxylase gene in brain cells and thus to influence catecholaminergic biosynthesis [23]. Catecholamines like DOPA, dopamine (DA), noradrenaline and adrenaline are essential neurotransmitters with important roles in brain diseases, exemplified by the depletion of DA being a key factor in PD [52,106,124]. Especially relevant to PD is that the enzyme tyrosine hydroxylase catalyses the rate-limiting step of DA synthesis [8]. Further research on BA's role as HDACI revealed protective effects for dopaminergic cells, namely rescuing them from αSyn-mediated DNA damage [125] or MPP+-induced toxicity [126] through an enhanced expression of DNA damage response genes. Supporting evidence come from a study in a drosophila model of PD in which BA has been reported to alleviate motor dysfunction and mortality [127]. Moreover, altered gut levels of SCFAs and neurotransmitters were associated with the surface area of the insula [9], a brain region that is understood to be dysfunctional in neurological and psychiatric disorders [128].

The influence of gut microbiota on PD might further impact the conventional therapy of levodopa administration, since the abundance of the gene for tyrosine decarboxylase, an enzyme converting levodopa to DA, in the microbiome of PD patients correlates with higher dosage needs for levodopa/carbidopa. Furthermore, it was shown in rats, serum levels of the aforementioned drug negatively correlated with the host's microbiome tyrosine decarboxylase gene levels [129]. These findings might provide the base for further clinical studies on gut microbial modulations in PD patients with increased levodopa/carbidopa dosages.

Taken together, SCFAs seem to exert overall beneficial effects on the CNS regarding autoimmune brain diseases and neurodegenerative diseases. However, preclinical findings on probable detrimental effects upon SCFA exposure in rodents suggests that these bacterial metabolites might function as double-edged swords when it comes to brain health. Thus, the thorough examination of these mechanisms is crucial before future potential therapeutic and preventative strategies can be unequivocally suggested.


**Table 1.** SCFA level alterations in brain diseases found in human studies.

This table shows the differences of SCFA levels of various sample materials from human patients compared to healthy controls. Significance of the data is given in the last column. ↑ symbolizes increased, ↓ decreased, whereas - symbolizes no significant change in metabolite levels found in cohorts with the specific disease. Sample material is noted as f *=* faecal; s *=* serum; p *=* plasma; u *=* urine with the associated reference as numbers in brackets; *p*-values < 0.05 are marked in bold letters. BA *=* butyric acid; AA *=* acetic acid; PP *=* propionic acid; VA *=* valeric acid; ASD = autism spectrum disorder; MDD = major depressive disorder; MS = multiple sclerosis; PD = Parkinson's disease. "\*" marked references are sourced from reviews.


**Table 2.** Gut-residing bacteria found to correlate with the production of SCFAs.

tSCFA *=* total SCFAs; BA *=* butyric acid; AA *=* acetic acid; PP *=* propionic acid; VA *=* valeric acid; HA *=* hexanoic acid. References are represented by numbers in brackets.

#### **Table 3.** Prevalence of dementia linked with various factors.


Multivariable logistic regression analysis models linking the prevalence of dementia and various factors from Saji et al. [93]. <sup>a</sup> Model 1: inclusion of enterotype I, <sup>b</sup> Model 2: Inclusion of enterotype III. Abbreviations: ApoE ε4 = apolipoprotein ε4; SLI = silent lacunar infarct; VSRAD = voxel-based specific regional analysis system for Alzheimer's disease.

#### **4. Non-SCFA Bacterial Metabolites**

The vast majority of current studies on the GBA involve SCFAs. Our gut microbiota, however, produces metabolites far beyond the products of fibre degradation, including

vitamins, polyphenol metabolites and products from amino acid metabolism (Figure 1). Each of these families of compounds are involved in various pathways and contain potential neuroactive metabolites [22]. This warrants our curiosity in exploring non-SCFA bacterial metabolites as contributors to the GBA.

#### *4.1. Amino Acid Metabolites*

Metagenomic studies suggests human gut microbes to be largely involved in amino acid metabolism [133]. Of special interest are the aromatic amino acids (AAA) tyrosine (Tyr), phenylalanine (Phe) and tryptophan (Trp). Humans are unable to produce AAA and depend on dietary sources and our gut microbiome for covering their nutritional needs. Gut bacteria are able to synthesize all three AAA de novo via the shikimate pathway [92,134]. In a first step, Trp and Phe are biosynthesized. Tyr is then synthetized from Phe. Further AAA metabolism occurs in the host as well as in gut microbes like *Lactobacillus*, Enterobacteriaceae and anaerobes of the phylum Firmicutes, that generate other metabolites. Phe and Tyr are catabolized in animals to neurotransmitters, including L-Dopa, DA, epinephrine and norepinephrine, while gut bacteria are able to produce phenolic compounds like p-cresol from Tyr and phenyl molecules from Phe. Trp is an essential precursor for the neurotransmitters serotonin and tryptamine, as well as vitamin B3 (niacin), redox cofactors NAD(P)+, plus metabolites from the kynurenine pathway [134]. On note, the kynurenine pathway in gut microbes generate metabolites associated with brain functions like indole, indole-derivatives, kynuric acid and quinolinate, which will be elaborated on in the following chapters. For a more in-depth analysis of AAA metabolism in plants, microbes as well as mammals, we refer the reader to the extensive review by Parthasarathy et al. [134].

Considering the previously described processes in AAA metabolism, it is conceivable to assume that gut microbiota might modulate neurotransmitter metabolism, synthesis, and availability in the gut, the circulatory system and the CNS. In fact, the abundance of circulating Trp can be curbed as a result of gut microbial Trp metabolization through other pathways, thereby possibly limiting the precursor for neurotransmitter synthesis in the CNS while also generating other neuroactive metabolites like indole and its derivatives [135] (Figure 3). On the other hand, gut microbes seem to elevate serotonin plasma availability after colonizing GF animals, leading to the assumption that the presence of a functioning gut microbiome contributes to physiological serotonin plasma levels [136]. More importantly, a recent study observed gut microbial involvements in Trp metabolism, providing an extensive overview of six pathways, each generating neuroactive metabolites referred to as "TRYP-6", consisting of kynurenine, quinolinate, indole, indole acetic acid (IAA), indole propionic acid (IPA) and tryptamine [135]. They identified five common gutinhabiting phyla capable of two to six pathways. The five phyla Actinobacteria, Firmicutes, Proteobacteria, Bacteroidetes and Fusobacteria thus have been suggested to relevantly influence Trp metabolism. Investigations on a genus level revealed that *Clostridium*, *Burkholderia*, *Pseudomonas*, *Streptomyces* and *Bacillus* were particularly capable of generating neuroactive Trp metabolites, with the first two holding the highest potential (Table 4). Numerous pathways and metabolites in the AAA metabolism, especially Trp, show relevant effects on the CNS that seem to be intricately complex and crucial for proper brain function, thus pointing to these non-SCFAs as promising players on the GBA (Figure 3).

: Parkinson's disease , Alzheimer's disease − − ↑ ↓ **Figure 3.** Hypothetical influences on brain diseases by gut bacteria-derived tyrosine and tryptophane metabolites. This figure illustrates the mechanistic effects by which gut microbial metabolites might influence brain functions related to autism spectrum disorder (ASD) and neurodegenerative disorders (NDs): Parkinson's disease (PD), Alzheimer's disease (AD), multiple sclerosis (MS) and psychiatric disorders (PsyD). Gut bacteria taking part in metabolite production are listed in black boxes situated under the orange (tyrosine metabolites) and blue boxes (tryptophane metabolites). Arrows accompanied with + or − represent an agonistic (+) or antagonistic (−) effect on a receptor, whereas unaccompanied arrows symbolize an effect described in the white boxes. ↑ = upregulated, ↓ = downregulated or lowered levels of. ROS = reactive oxygen species; NMDA-R = N-methyl-D-aspartate receptor; Trp = tryptophane; AHR = aryl hydrocarbon receptor; 4EPS = 4-ethylphenylsulfate; IPA = indole-propionic acid; IAA = indole-acetic acid; kyn = kynurenine; quin = quinolinate; trypt = tryptamine; i.a. = inter alia.

#### 4.1.1. AAMs and Neurodevelopmental Disorders

P-cresol is a known uremic toxin, which is metabolized into p-cresol sulphate by the liver [137] and is believed to derive from Tyr fermentation in several gut bacterial species (Table 4). Significantly increased urinary and faecal levels of p-cresol were reported in autistic children, with some linking urinary levels with the clinical severity of disease [39,41,138,139]. Interestingly, p-cresol levels significantly and negatively correlate with age in ASD patients, which might suggest that younger individuals with ASD are exposed to effects from elevated p-cresol levels [41]. One study, however, did not detect

significantly altered faecal levels in children with ASD [42]. As for non-human studies, p-cresol was very recently shown to dose-dependently induce and exacerbate ASD-like behaviours and significantly activate dopamine (DA) turnover in brain regions (amygdala, nucleus accumbens and striatum) in the genetically vulnerable BTBR mice model for ASD [140]. Social avoidance behaviour and increased gut levels of p-cresol were detected in GF mice, inoculated with p-cresol-producing Clostridiales (including Lachnospiraceae and Ruminococcae families), and these mice associated with defective myelination in the prefrontal cortex [141]. Additional in vitro testing showed that exposure to p-cresol interrupted the differentiation of progenitor cells into oligodendrocytes [141], suggesting that gut microbial p-cresol might impact CNS myelination through transcriptional changes. Other mechanisms by which p-cresol might negatively impact neuronal functions [140] include the inhibition of dopamine-β-hydroxylase and membrane depolarization with higher vulnerability for seizures and blunted Na+/K+-ATPase function. These mechanisms might demonstrate a potential for gut bacterial-derived p-cresol to play a role in disorders with disfunctions in the CNS, including ASD, MS, and neurodegenerative diseases.

In the maternal immune-activated (MIA) mouse model of autism spectrum disorder, changes in serum metabolites, showing significant elevations of two AAA bacterial metabolites 4-ethylphenylsulfate (4EPS) and indolepyruvate were found, which were completely normalized along with ASD-related behaviour, dysbiosis and impaired gut barrier after the inoculation with the probiotic *B. fragilis* [142]. Moreover, WT mice treated with the metabolite 4EPS alone manifested anxiety-like behaviour similar to MIA-mice, thus suggesting a compelling association between 4EPS and ASD. Additionally, other metabolites, two of them being serotonin and p-cresol, were elevated in the serum of MIA-mice, though not at significant levels [142]. It is essential to mention 4EPS's structural similarity to the prior mentioned p-cresol, which has links to ASD and is believed to share its producers in the gut with 4EPS, namely *Clostridia* spp. [36,140,142] (Table 4). Overall, preclinical data and some supportive human studies point towards a connection between 4EPS, indolepyruvate, p-cresol and ASD, not only as biomarkers of disease but also as putative mediators of pathogenesis.

With an extensive in silico study, Kaur et al. [135] recently detected the aforementioned "TRYP6", the six Trp metabolism pathways generating neuroactive metabolites, to be enriched in the metagenome of autistic gut microbiota. Genomes of the genera *Burkholderia* and *Pseudomonas* showed particularly large potentials for TRYP6 metabolism. *Burkholderia* holds pathways for kynurenine and quinolinate, with a lower production of IAA, indole and tryptamine, while *Pseudomonas* is a strong producer for kynurenine and a weaker one for IAA, quinolinate and tryptamine. Other enriched pathways in autistic children consisted of those generating indole and its derivative IAA by already mentioned genera *Burkholderia* and *Pseudomonas* plus *Corynebacterium*. Though microbiota from NT individuals also harboured some relatively enriched bacteria capable of producing TRYP6, namely *Alistipes* for indole and *Eggerthella* for IAA production, these genera are comparatively weak producers and thus, theorized to use indole and IAA as inter-bacterial communication tools [135]. Similarly, altered Trp metabolism in ASD has been indicated through significantly increased urinary levels of IAA, indoxyl sulphate (IS, also known as indican) and indolyl lactate in autistic children [143], though to date no data on faecal levels have been found.

In addition to Trp metabolism, altered Phe metabolism in autistic children were recently highlighted, partly based on evidence of significantly elevated Clostridia-generated Phe metabolites in the urinary profiles of ASD patients including 3-(3-hydroxyphenyl)- 3-hydroxypropionic acid, 3-hydroxyphenylacetic acid and 3-hydroxyhippuric acid [36]. However, whether these metabolites might be modulated by gut bacteria remains to be elucidated.

Increased faecal levels of glutamate in children with ASD, as well as decreased GABA levels in those with pervasive developmental disorder not otherwise specified (PDD-NOS) have been previously reported [39]. Kang et al. [41] have similarly detected lower GABA

faecal levels in autistic children, though these did not reach significance (*p* = 0.077). Preclinical observations in altered GABA and glutamate levels were made in a study invoking behaviours associated with ASD in GF mice through inoculation with gut microbiota derived from autistic patients. In comparison, mice receiving healthy donor FMT did not produce ASD-like behaviour [45]. Moreover, lower faecal levels of the GABA A receptor agonists 5AV and taurine were found in the first group, supporting a putative role of disturbed GABA signalling in ASD. In a further step, the exposure of an ASD mouse model to taurine or 5AV during the prenatal and weaning period produced mice with ameliorated ASD behaviour in comparison with mice treated during their juvenile stage and older mice. This suggests a critical window of vulnerability for disturbed GABA signalling during neurodevelopment [45]. Other researchers were able to uncover correlations between gut microbe genes associated with neurotransmitter metabolism and the surface area of the insula, with a focus on two microbial genes involved in GABA and glutamate metabolism, namely 4-hydroxybutyrate dehydrogenase and glutamate dehydrogenase [9]. As already mentioned, the insula is thought to be dysfunctional in many psychiatric disorders with disturbed emotion, cognition and motivation, such as affective, neurodevelopmental and neurodegenerative disorders [128]. However, it is also important to note that Kang et al. [41] were not able to detect any significant changes in gut bacterial pathways by PICRUSt database analysis between ASD and NT children. Nevertheless, these studies accumulatively show the potential involvement of gut microbial metabolites in disturbed GABA and glutamate signalling in ASD pathophysiology. Further metabolomic, metagenomic and microbial analyses of faecal amino acid metabolites (AAMs) are nonetheless highly encouraged. Compellingly, a recent study reported that GABA produced from gut bacteria (*E. coli HT115* and *P. aeruginosa PAO1*) was able to protect from neurodegeneration in the nematode C. elegans [144].

Lastly, the amplified metabolism of the amino acids Tyr, lysine, cysteine and methionine in healthy children's gut microbiomes have been found, which implies a supportive function of gut commensals during brain development, since the mentioned amino acids are not only substrates for the synthesis of structural proteins, but also neurotransmitters and biogenic amines [145].

#### 4.1.2. AAMs and Psychiatric Disorders

No specific studies on non-SCFA faecal metabolites have been found in humans with affective disorders. However, there are a handful of reports pointing to correlations between disturbed gut microbial Trp metabolism and psychiatric disorders like anxiety and depression. Studies on acute tryptophan depletion (ATD) in humans demonstrated a correlation between the reduced levels of circulating Trp and depressive symptoms in patients, who are responsive to treatment with selective serotonin reuptake inhibitors [146,147]. ATD has also been shown to worsen depressive symptoms in patients in remission, as well as in healthy volunteers at high risk for depression [70]. Furthermore, 5-HT levels in the CNS were shown to be impacted by the amount of dietary Trp in humans [147]. This raises the question of whether a disturbed gut microbial Trp metabolism could deplete circulating Trp availability and consecutively impact 5-HT homeostasis in the CNS. The impact of gut microbiota, or the lack thereof, on the host's nervous system can be explored in GF raised animals providing the evidence for altered levels of neurotransmitters in the brain in comparison to conventionally raised control animals [148]. In support of the previous reports on humans, a recent study in GF mice showed initially higher Trp and 5-HT brain levels together with a less depressive behaviour at baseline and intriguingly, decreased Trp and 5-HT with enhanced depressive behaviour after ATD compared to the control group (specific pathogen-free mice) [149]. Additional support was obtained by findings in a rat model, which showed an induced depression to invoke gut microbial alterations as well as noticeable faecal metabolite shifts [150]. Sixteen metabolites were evaluated to be significantly distinct enough to function as depression biomarkers, including altered Trp metabolites (upregulated dextrorphan O glucuronide, 3-methyldioxyindole and down-

regulated 5-methoxytryptophan) (Table 5). The others consisted of bile acid metabolites (upregulated) as well as hypoxanthine (upregulated) and fatty acid metabolites (downregulated). Additionally, the altered gut microbiota also resulted in changes of catecholamine levels in the hippocampus of depressed rats, specifically serotonin (5-HT) and DA [150].

Similarly, Clarke et al. [151] found elevated levels of 5-HT and 5-HIAA in hippocampal structures of GF male mice, as well as higher plasma levels of their precursor Trp. Considering that CNS Trp levels are to a great extent regulated by its abundance in plasma, this supports the conjecture of a humoral pathway through which gut microbes could influence serotoninergic neurotransmission by modulating Trp availability [148]. Fascinatingly, it was not possible to reinstate altered hippocampal 5-HT levels through inoculation with an intestinal microbiota in GF mice at a later stage of life, even though the serum levels of Trp were normalized. This points to a critical time window in which microbial Trp metabolism could directly impact neurodevelopment [152]. Additionally, GF animals displayed elevated stress reactivity, represented with higher corticosterone production, while also expressing lower anxiety-like behaviour that was normalized after recolonization [148]. This is intriguing, since stress hormones like cortisone shift Trp metabolism away from 5-HT production to the kynurenine pathway that generates kynurenic acid, quinolinic acid and picolinic acid [153]. While kynurenic acid invokes antagonistic effects on the α7 nicotinic acetylcholine receptor and the N-methyl-D-aspartate (NMDA) receptor, quinolinic as well as picolinic acids are agonists of the NMDA receptor with neurotoxic and depression-producing properties [152]. It is also noteworthy, that a previously mentioned study on ASD reported the induction of anxiety-like behaviour in mice following the administration of the microbial Trp metabolite 4EPS [142]. Overall, these studies might provide tangible evidence for gut microbial impact on depressive and anxious behaviour by regulating the availability of circulating Trp and consecutively, levels of Trp metabolites like 5-HT, kynurenic, quinolinic and picolinic acids. However, clear associations between gut microbial Trp metabolites and depression or anxiety seem too early to be made since studies on this subject are largely based on preclinical settings on animals.

Indole is partly produced from dietary tryptophan through the enzyme tryptophanase [154] mainly by gut bacteria *Escherichia*, *Citrobacter*, *Fusobacterium*, *Bacteroides*, *Clostridium\_XIX*, *Desulfitobacterium*, *Edwardsiella*, *Providencia* and *Shigella* [135] (Table 4). Indole has also recently been shown to be associated with impaired motor activity, anxiety and depression in rats when acutely or chronically overproduced [155]. Probable pathways by which these effects occur are the activation of vagal afferences by indole and on the other hand, accumulation of oxidized indole derivatives like oxindole and isatin in the brain. Indole has been shown to activate gut mucosal L-cells to secrete glucagon-like peptide-1 (GLP-1), which then stimulates vagal afferent fibres, therefore presenting an indirect impact of indole on the CNS [156]. Oxindole is known to inhibit motor activity [155], invoke hypotension, loss of righting reflex and a reversible comatose state, while isatin is proposed an anxiogenic role by inhibiting monoamine oxidase (MAO) B and by producing antagonistic effects on benzodiazepine receptors in rodents [70]. However, Jaglin et al. [155] showed that while acute overexposure to indole in the rat gut produced depressant effects on motor activity and elevated levels of oxindole and isatin the brain, chronic exposure to indole-producing *E. coli* induced depression-associated traits (anxietylike and helplessness behaviours) without an accumulation of oxindole and isatin in the CNS. This suggests that an indole-overproducing gut microbiome might be a risk factor for the development of anxiety and depression, while acute spikes of indole-production might profoundly decrease locomotion by the central accumulation of oxindole and isatin as well as activation of vagal afferences. Studies have further demonstrated vagal GBA connections to neurons related to reward centres, thus pointing towards a probable pathway for gut metabolites to influence the brain in neuropsychiatric disorders with disturbed reward systems [157]. A recent in silico study on microbial Trp metabolism pathways in neurological disorders called for further investigations of the gut microbiome in schizophrenia, since assumptions were too early to be made on one single available dataset, that showed altered

indole, IAA and tryptamine pathways in the microbiome of schizophrenic patients [135]. The research of gut microbial influence on schizophrenia is still in its infancy, which is represented by very few analyses on microbiome compositions and no study on faecal metabolomes in schizophrenic cohorts so far. With the emerging correlations between other brain disorders and the microbiome, as well as some preclinical information on dysbiosis and probiotic studies in schizophrenia [158], further work on this particular subject remains wanting of exploration. Taking all these studies into consideration, tryptophan metabolism with its manifold metabolites seem to be intricately influenced by gut microbial metabolites and to be implicated in psychiatric disorders and brain functions.

#### 4.1.3. AAMs and Neurodegenerative Diseases Alzheimer's Disease

A limited number of studies have indicated adverse effects on neurons in the context of Alzheimer's disease (AD) by Trp metabolites. The decarboxylated molecule tryptamine has been associated with neurotoxicity and neurodegeneration [92,159–161] (Figure 2). Tryptamine producers commonly found in gut flora are *Holdemania*, *Desulfovibrio*, *Yersinia*, *Tyzzerella*, *Bacillus*, *Clostridium* and *Ruminococcus* [135] (Table 4). Most recent findings showed gut bacterial genomes in faecal samples of AD patients, of which one gene sequence encodes the enzyme Na-transporting NADH:Ubiquinone reductase (in *Clostridium sp.*), which produces the neuroprotectant ubiquinone. Interestingly, that enzyme is also involved in the metabolic synthesis of AAA [92]. Underlining these findings in AAA metabolism, Trp and Tyr (also GABA, taurine and valine) were found to be decreased in faecal samples of mice receiving FMT from an AD patient [97], though as noted in a previous chapter, two faecal donor samples selected out of groups of 14 healthy and 13 AD volunteers might limit the evidential impact by probable inter-individual variations.

Regarding the theory of a perturbed Trp metabolism in AD, microbial and hepatic enzymes generate kynurenine from Trp, and in succession, kynurenic acid or quinolinate. Quinolinate shows excitotoxic properties as an NMDA receptor agonist, whereas kynurenic acid ameliorates those neurotoxic effects as NMDA receptor antagonist. Therefore, this might provide a probable link between Trp metabolites and neurodegenerative processes in AD (Figure 3). In contrast to those findings, in aberrantly elevated amounts, kynurenic acid has been linked to cognitive impairments, probably caused by its antagonistic effect on the α7-nicotinic acetylcholine receptor [152]. It should be emphasized that CNS kynurenine mostly originate from the periphery and that its metabolization into kynurenic acid and quinolinate takes place in the CNS [151]. Some gut genera *Bacillus*, *Burkholderia*, *Streptomyces* and *Pseudomonas* are specially equipped for kynurenine production, while *Klebsiella*, *Bacillus* and *Burkholderia* are efficient quinolinate producers [135].

Another Trp metabolite generated from gut microbiota is indoxyl sulphate (IS), an uremic retention toxin in patients with chronic kidney disease, which has been associated with cognitive impairments [162] and various diseases of the brain such as AD, PD, MDD and MS [163]. IS was previously observed to induce nuclear factor-kappaB (NF-κB)-mediated oxidative stress in animal and in vitro studies [163,164]. Moreover, it exhibited potential neurotoxic effects in mice through perturbed microglial and astrocyte function, resulting in neuronal death [165]. This might be of interest since oxidative stress is proposed as a major process in neurodegenerative diseases [166]. Underscoring these correlations, researchers observed an elevated cerebrospinal fluid (CSF)/plasma ratio of IS in patients with PD compared to healthy counterparts [163]. This might suggest the increased crossing of IS through the BBB, a process probably facilitated by increased BBB permeability in diseases like AD and PD [167]. On the other hand, decreased IS levels in CSF, serum and faecal samples of GF and AT mice have been associated with perturbed fear extinction learning processes. These defects are common in anxiety and fear-related diseases with impaired learning and memory [168]. Considering these early preclinical and at times inconsistent findings on IS and brain disorders, future research focusing on microbiome-derived IS and its participation in neurodegenerative processes might enlighten this complicated and

emerging subject. In a dementia-prone mice model, faecal metabolites seemed to differentiate from HC through higher levels of the amino acids ornithine and Tyr, which might excite further research, considering Tyr's role as precursor for several crucial neurotransmitters (norepinephrine, epinephrine and DA) as well as for the uremic toxin p-cresol. Moreover, ornithine has shown protective effects in neurotoxic ammonia [114,169], though it remains to be determined whether bacterial metabolism is involved. Nevertheless, these reports overall suggest gut microbial modulated AAA metabolites as potential components of the complex and emerging field discussing the influence of GBA in AD.

#### Parkinson's Disease

The analysis of gut microbial Trp metabolism in several databases of PD by Kaur et al. [135] detected enriched indole pathways and three of its producers to be differentially abundant in PD, namely *Alistipes*, *Akkermansia* and *Porphyromonas*. Concomitantly, enriched IAA production pathways in combination with increased *Lactobacillus* and *Staphylococcus* abundances were measured. Interestingly, distinct alterations of kynurenine and quinolinate pathways were undetectable, unlike in other neuropsychiatric disorders like ASD [135]. Congruously to these activated production pathways of IAA, increased IAA urinary levels were reported in patients with idiopathic PD [170,171] (Table 6). However, decreased serum IAA were observed in two cohorts of Japanese patients with idiopathic and familial PARK2-mutated PD [172,173], thus showing some inconsistencies across human studies. Some gut bacteria capable of generating IAA are *Klebsiella*, *Ralstonia*, *Staphylococcus*, *Bacillus*, Clostridia, *Bacteroides* and *Escherichia* [135,174] (Table 4). Interestingly, IAA was previously mentioned to suppress pro-inflammatory cytokine production by macrophages and act on the aryl hydrogen receptor (AHR) [109]. IAA was also able to attenuate neuroinflammation in LPS-stimulated BV2 microglia in vitro [175]. Overall, these findings point towards a probable role of altered gut microbial production of IAA in neuroinflammatory processes in PD, even if further investigations need to disentangle the complexities between gut-derived IAA and PD.

The previously mentioned metabolite and uremic toxin p-cresol (or its hepatically sulfonated form p-cresol sulphate) in ASD [142] is generated through intestinal bacterial Tyr metabolization [134], with especially strong producers within Coriobacteriaceae and Clostridium clusters XI and XIVa [137]. P-cresol sulphate has been previously associated with neurological impairments in chronic kidney disease [162]. Moreover, two recent studies reported significantly higher p-cresol sulphate levels in the CSF (yet not in plasma) from PD patients compared to samples from HC [163,176] (Table 6). Additionally, higher CSF to plasma ratios in PD was observed in one study, suggesting that individuals with PD accumulate more p-cresol sulphate in the brain than their healthy counterparts [163,176]. These findings support the relevance of a perturbed BBB allowing the increased permeation of putative neurotoxic microbial metabolites from circulation to the brain in PD [165]. Moreover, p-cresol sulphate levels in CSF associated with the presence of motor fluctuations in PD patients, suggesting a correlative connection with disease progression [163]. This is supported by the fact that p-cresol is a known inhibitor of dopamine–beta-hydroxylase [177], the enzyme facilitating the conversion from DA to norepinephrine. Therefore, alterations in the p-cresol production of the microbiome, as well as the gut bacterial impact on BBB integrity, might regulate neurotransmitter metabolism in the brain. Cirstea et al. [132] have recently provided further evidence that associates altered gut microbial metabolism, disturbed gut function (constipation and IBS) and PD. Compared to HC, PD patients harboured decreased levels of common BA-producing Clostridia, including some Lachnospiraceae genera (*Roseburia*, *Coprococcus)* and *Faecalibacterium*), as well as enriched bacterial clusters associated with p-cresol and phenylacetylglutamine production (*Christensenellaceae*, *Ruminococca*, *Akkermansia*, *Oscillospira*, *Mogibacteriaceae*). Moreover, increased serum levels of p-cresol and phenylacetylglutamine were measured, showing positive correlations with the presence of PD, as well as the severity of gut dysfunction. Therefore, a gut microbiome shift from BA producers to microbes generating

AAMs such as p-cresol and phenylacetylglutamine might influence symptoms of intestinal dysfunction, as well as altered circulating metabolites in patients with PD.

Manganese (Mn) has been shown to evoke neurodegenerative processes when accumulated in the brain in the context of PD [178,179]. Recent findings discovered Mn exposure to alter gut bacterial genes involved in amino acid and neurotransmitter metabolism (GABA, glycine, glutamate, Trp, Phe) in a mice model, thus giving rise to the novel proposition of gut bacterial involvement in manganese-associated neurotoxicity [180].

These findings overall suggest that members from the Clostridia class seem to be implicated in the production of AAA metabolites (phenolic and indole derivatives) associated with neurodegenerative disorders. Since Clostridia are also known as key BA producers of the human gut [73], and as the chapters above have discussed, a probable beneficial connection between BA and brain diseases, further investigations in Clostridia-derived metabolites and various brain diseases seem warranted to elucidate the relevance of these bacteria. Furthermore, our collected data on indole and indole derivatives show correlations with various brain disorders (autism, anxiety, PD) that are often accompanied with gut issues [36,110,132,181].

#### 4.1.4. AAMs and Autoimmune Diseases of the Brain

As mentioned above, bacterial production of indole from dietary Trp might be involved in perturbed brain functions. The extensive study by Rothhammer et al. [182] has suggested, that in combination with type I interferons (IFN-Is), gut bacterial-derived metabolites might suppress neuroinflammation through agonistic effects on the aryl hydrogen receptor (AHR) on astrocytes (Figure 3). Serum levels of the AHR agonists indole, indoxyl-(3-)sulphate (IS), IPA and indole-3-aldehyde were found to be lower in patients with MS than in HC (Table 6). Furthermore, experiments conducted in EAE mice models of MS uncovered that depleted dietary Trp exacerbated disease, while the administration of the AHR agonists IS, IPA or indole-3-aldehyde reduced disease burden [182]. IS specifically was further shown to cross the BBB and to stimulate AHR on astrocytes. Not unrelatedly, another bacterial indole derivative, IAA, was able to attenuate neuroinflammation in LPS-stimulated BV2 microglia in vitro [175], which underlines the hypothesis, that bacterial-derived indoles might benefit neuroinflammatory processes by activating the AHR on brain cells. It might further be worth noting that the Trp metabolite and indole derivative IPA was shown to cross the BBB and to ameliorate harmful reactive oxygen species (ROS) in the brain as a neuroprotectant [135]. Described gut genera capable of generating IPA are few and belong to the Firmicutes phyla, namely *Clostridium*, *Peptostreptococcus*, *Escherichia* and *Proteus* [134,135] (Table 4). Therefore, the presence of these IPA-producing genera may indicate beneficial anti-oxidative properties for brain function.

Interestingly, p-cresol producing Clostridiales (including Lachnospiraceae and Ruminococcae families), appeared to be abundant in MS patients' microbiomes [79,183], which might potentially lead to similar detrimental effects on the CNS as discussed in the chapters on ASD and PD, though no reports on elevated p-cresol in MS patients exist as of now. The only MS study in humans to investigate the faecal metabolome, as far as our literature search was able to capture, provided no relevant findings on non-SCFA bacterial metabolites [29,81]. Future research is encouraged to further develop and confirm these initial findings by conducting metabolomic, gut taxonomical and metagenomic tests in the faecal samples of MS patients in order to look for correlations with bacterial metabolites beyond SCFAs.


**Table 4.** Gut-residing bacteria found to correlate with the production of amino acid metabolites (AAMs).

GABA = gamma-aminobutyric acid; TRYP6 = six Trp metabolism pathways generating the neuroactive metabolites in brackets ("); 4EPS = 4-ethylphenylsulfate; IAA = indole-acetic acid; IPA = indole-propionic acid. References are listed as numbers in brackets.

**Table 5.** Sixteen faecal metabolites with significant correlation in rats with induced depression (adjusted from Yu et al. [150]).


**Table 5.** *Cont.*


Faecal levels in depressed rats compared to healthy control group: ↑ = upregulated; ↓ = downregulated. Rows marked in blue have a *p*-value of *p* < 0.01, whereas rows in white reached *p* < 0.05. MG = monoacylglyceride; PE = phosphatidylethanolamine; PS = phosphatidylserine

**Table 6.** Bacterial AAMs correlated with brain diseases or the progression of brain diseases.


This table shows the differences of AAM levels of various sample materials from human patients compared to healthy controls. *p*-values are listed in the last column while the significant date is written in bold letters. ↑ symbolizes increased, ↓ decreased, whereas—symbolizes no change in the metabolite levels found in cohorts with the specific disease. Sample material is noted as f = faecal; s = serum; c = cerebrospinal fluid; p = plasma; u = urine; s/p = blood samples with the associated reference as numbers in brackets. References noted with "\*" are sourced from reviews. *p*-values < 0.05 are marked in bold letters. <sup>a</sup> correlated with progression of disease. <sup>b</sup> (*p* = 0.00364 PD in early stage of disease, *p* < 0.001 PD in mid-stage, *p* = 0.056 PD in late-stage compared to HC). (For example: ccap <sup>a</sup> = altered cerebrospinal fluid sample, and altered cerebrospinal as well as plasma samples which correlate with disease progression). IAA = indoleacetic acid; IS = indoxyl sulphate; GABA = gamma-aminobutyric acid; IPA = indole propionic acid.

#### *4.2. Other Metabolites*

#### 4.2.1. Trimethylamine N Oxide (TMAO)

The gut bacterial fermentation of dietary L-carnitine and phosphatidylcholine, which are abundant in red meat, produces trimethylamine (TMA). The following hepatic oxidization by flavin-containing monooxygenase 1 and 3 (FMO1 and FMO3) produces trimethylamine N oxide (TMAO), a metabolite frequently linked to increased risk of cardiovascular, metabolic and cerebrovascular disease, whether as a mediating factor, marker or bystander of disease [185–188]. Recent studies have detected TMAO in human cerebrospinal fluid (CSF), thus establishing its presence in the brain beyond the cerebrovascular system [188,189]. It is to mention that TMAO plasma levels are subject to factors besides gut microbiome composition, namely diet and liver enzyme activity. It is presently unclear how much circulating levels in the CSF depend on the de novo biosynthesis of TMAO in the brain. Recent studies have shown, however, strong correlations between CSF and plasma levels, suggesting that TMAO brain-levels largely derive from the availability in blood, thus supporting the theory of peripheral TMAO reaching the CNS [163]. Interestingly, TMAO plasma levels were found to be significantly higher in elderly humans as well as in aged mice compared to their respective younger groups [190–193]. This would be coherent with recently reported associations between shifts in gut microbiota and presence of neurodegenerative disorders [89,110,194].

By assessing TMAO CSF levels in volunteers with AD dementia, mild cognitive impairment (MCI) and healthy controls (HC), as well as correlations between TMAO CSF levels and biomarkers of AD and neurodegeneration, Vogt et al. [188] have reported the potential involvement of TMAO in AD. They found significantly higher CSF levels of TMAO in the AD and MCI groups compared to HC, with no differences between AD and MCI [195], all while controlling for age, sex and APOE ε4 genotype. Moreover, they discovered CSF TMAO levels to be significantly correlated with AD biomarkers that indicate a connection to tau pathology and axonal injury. Congruously, a different study observed plasma TMAO levels to be inversely correlated with cognitive functions (working memory, episodic memory and fluid cognition) in middle-aged to older adults [193].

Conversely, a recent study reported no differences in TMAO levels between CSF samples from PD patients and HCs. Nevertheless, significant TMAO elevations were detected in PD patients with motor fluctuations compared to those without (Table 7), thus pointing to a role of TMAO in disease progression [163].


**Table 7.** Other bacterial metabolites correlated with brain diseases or progression in humans.

AD = Alzheimer's disease; MCI = mild cognitive impairment; PD = Parkinson's disease, ASD = autism spectrum disorder; TMAO = trimethylamine N oxide, <sup>a</sup> correlated with the progression of disease. Significance of data is *p* ≤ 0.05. ↑ symbolizes increased metabolite levels found in cohorts with the specific disease compared to healthy controls. Sample material is noted as f = faecal, c = cerebrospinal fluid, p = plasma sample with the associated reference as numbers in brackets.

Researchers have previously mentioned shared pathological arteriosclerotic and inflammatory mechanisms between cardiovascular and dementia-associated cerebrovascular diseases [93]. Literature on vascular cognitive impairment (VCI), a broad definition encompassing cognitive disorders with associations to any kind of cerebral vascular brain injury, further proposed that risk factors for VCI, including hypertension, hypercholesterolemia, diabetes mellitus, atrial fibrillation and others, might overlap the ones for AD [196]. Additionally, results in genetically modified mice previously indicated TMAO to cause the progression of atherosclerosis, a risk factor for dementia [195,197]. TMAO was further implicated with decreased reverse cholesterol transport in mice [198] and enhanced platelet

hyperreactivity and thrombosis risk in mice and human subjects [199]. This might suggest a vascular aspect of the mechanism by which this bacterial metabolite might take part in the pathophysiology of AD. However, as Vogt et al. [188] found, differences in TMAO levels between healthy controls, MCI and AD groups were independent from traditional cardiovascular disease risk factors such as body mass index, blood pressure, cholesterol and fasting glucose. Positive associations between TMAO levels and biomarkers for AD and neurodegeneration were also further controlled for peripheral vascular disease risks factors, thus implying that TMAO might affect neurodegeneration by other means than vascular mechanisms. Overall, whether and to what extent TMAO might influence AD by the promotion of vascular disfunction is still to be determined at this point in time.

In studies investigating TMAO's involvement in other mechanisms of neurological diseases, this bacterial metabolite was proposed to weaken the BBB by downregulating tight junction proteins in humans [200]. After reaching the CNS, TMAO was shown to promote neuronal senescence in the hippocampus and cognitive impairment in mice by increasing oxidative stress, disturbing mitochondrial dysfunction and inhibiting the mammalian target of rapamycin (mTOR) signalling, which increased synaptic damage as well as reduced synaptic plasticity-related proteins [190]. A study with the APP/PS1 transgenic mice model of AD indicated higher TMAO-levels in plasma to be associated with cognitive and pathological deterioration, while treatment with the TMA formation inhibitor 3,3-Dimethyl-1-butanol (DMB) alleviated cognitive deterioration and defective synaptic plasticity [191]. Furthermore, DMB treatment managed to reduce hippocampal neuroinflammation and AD-associated pathologies like Aβ42, β-secretase and βCTF (βsecretase-cleaved C-terminal fragment) levels in APP/PS1 mice [191]. Similarly, a different transgenic mouse model of AD (3x Tg-AD) displayed significantly elevated plasma and brain TMAO levels in comparison to healthy controls in a recent study by Govindarajulu et al. [192]. They further incubated hippocampal brain slices of wild-type mice with TMAO and found deficits in synaptic plasticity, impaired synaptic transmission, altered presynaptic and reduced postsynaptic glutamatergic receptor units, as well as induced endoplasmatic reticulum (ER)-mediated protein kinase RNA-like endoplasmic reticulum kinase (PERK) pathway [192]. Another study has further indicated TMAO to impair cognitive function by promoting neuroinflammation and astrocyte activation in mice [193]. In addition to the association between TMAO levels and neuroinflammation and astrocyte activation in older mice, TMAO supplementation in young mice for six months exhibited a decline in memory and learning (assessed through the novel object recognition test) and indeed, elevated markers for neuroinflammation and astrocyte activation. Furthermore, human astrocyte cultures incubated with TMAO showed altered cellular morphology and markers indicating astrocyte activation, thus proposing a direct effect of TMAO on astrocytes [193]. Overall, all of these preclinical findings suggest that TMAO may provoke cognitive impairment by promoting neuroinflammation, AD-related amyloid formation, oxidative stress, as well as deficits in synaptic plasticity and function by promoting ER stress-mediated PERK signalling pathways.

Additionally, an in silico study detected significant correlations between TMAOrelated genes and AD biomarkers in nine potential genetic pathways involved in both, that might underline the proposition for TMAO as a strong biomarker for AD [201]. Those nine pathways include in no specific order: the metabolism of proteins; immune system; adaptive immune system; Alzheimer's disease; axon guidance; amyotrophic lateral sclerosis (ALS); erythropoietin-producing human hepatocellular receptor A (EPHA) and B (EPHB) forward signalling and metabolism of lipids and lipoproteins. These findings might be used as groundwork for investigations of specific pathways that might elucidate a diet-microbial metabolite–brain disease axis.

Curiously, studies in mice and in vitro models reported disease-mediating as well as protective mechanisms by TMAO on processes in neurodegenerative disorders such as a reduction in amyloid aggregation in AD and PD, thus providing a new potential therapeutic target [22,200]. However, more congruous study results have been reviewed and reported on the pro-inflammatory effects of increased TMAO plasma concentrations [200]. These reports underline the need for further studies on TMAO and its combined effects on enhanced circulating pro-inflammatory mediators that can potentially cross a TMAOinduced disturbed BBB to a greater extent, and therefore promote the aforementioned neuroinflammatory and neurodegenerative effects in the brain.

Some gut bacteria found to be significantly associated with enhanced plasma TMAO were the genera *Prevotella*, *Mitsuokella*, *Fusobacterium*, *Desulfovibrio*, *Methanobrevibacter smithii*, and some from the Lachnospiraceae and Ruminococcaceae families [186] (Table 8). Three of those genera belong to the Bacteroidetes and six to the Firmicutes class, congruous with a recent study in dementia [93] that showed a significantly higher Firmicutes/Bacteroidetes ratios in demented individuals with MRI-detected silent lacunar infarctions, as well as strong correlations between dementia and low counts of *Bacteroides* along with higher counts of 'other bacteria'. Additional sources have reported the following TMA-producing genera: *Anaerococcus*, *Clostridium*, *Escherichia*, *Proteus*, *Providencia* and *Edwardsiella* [22]. Overall, future work is strongly advised to address and investigate TMAO brain levels and any correlations with the gut microbiome.

#### Carnitine Analogues

Given that gut metabolites related to carnitine such as TMAO affect the GBA, researchers have recently discovered two new potentially brain active metabolites, namely the carnitine analogues 3-methyl-4-(trimethylammonio)butanoate (3M4-TMAB) and 4- (trimethylammonio)pentanoate (4-TMAP) [202]. These compounds are generated by gut bacteria from the family Lachnospiraceae (*Clostridiales symbosium* and *clostridioforme*) (Table 8) and were absent in both the gut and brain of GF mice but present in controls, thus suggesting that gut microbiota may be responsible for their presence in the brain. Importantly, these compounds were found in identical regions of white matter of the brain as carnitine and showed an inhibition of carnitine-mediated fatty acid oxidation (FAO) in a murine cell culture model of CNS white matter [202]. FAO is crucial for neuronal energy homeostasis, and inborn errors of this system may link faulty neuronal stem cell self-renewal to ASD [203,204]. Considering these findings and that increased abundance of Clostridia have been linked to ASD and other neurodevelopmental disorders [205], faulty FAO promoted by bacterial metabolites might be an important topic to explore in the future.

#### 4.2.2. Polyphenolic Metabolites

Polyphenols derived from dietary sources are metabolized by human gut microbes to phenolic acids. Research has shown gut microbial metabolites of dietary polyphenols to accumulate in brain tissue and to modulate α-synuclein misfolding, aggregation and neurotoxicity in vitro and in an animal models [206]. Among those metabolites were 3-hydroxybenzoic acid (3-HBA), 3-(3-hydroxyphenyl)propionic acid (3-HPPA) and 3,4 dihydroxybenzoic acid (3,4-diHBA). Further investigations led to the detection of *B. ovatus* as a producer of the aforementioned metabolites that were converted from the dietary polyphenols (+)-catechin (C) and (−)-epicatechin (EC). C/EC-independent production of 3,4-diHBA, 3-HBA and dihydrocaffeic acid (DHCA), a circulating anti-inflammatory phenolic acid, also occurred through *B. ovatus*, *E. lenta* and *E. coli* (Table 8). In earlier studies, 3-HPPA and 3-HBA were shown to ameliorate Aβ-peptide misfolding [207], therefore underlining the potential of bacterial polyphenolic metabolites to protect from neurotoxic protein-aggregation and neurodegenerative disorders like AD and PD. Additional gut bacterial metabolism-derived polyphenols, particularly enerolactone and enterodiol, aryl-γ-valerolactone metabolites and urolithin A/B, exhibit polyphenolic neuroprotective properties [208]. Two studies have further summarized mechanisms by which brain function could be influenced by bioactive microbe-derived metabolites of polyphenols [209,210]. Direct neuroprotective impacts include modulating neuronal receptors, antioxidation, antiinflammation and overall neuroprotective effects. Indirect mechanisms encompass the modulation of gut microbial homeostasis by supporting beneficial bacteria while decreasing pathogens, as well as improving cerebrovascular health by increased nitrogen oxide levels and vasodilatory response. Limits of the summarized findings were put by their in vitro and ex vivo-based designs.

The ability to cross the BBB is crucial for metabolites to exert neuroactive properties and ten microbe-derived polyphenolic metabolites were found to be capable of distributing in rat brains after intravenous administration [211]: 4-hydroxyhippuric acid, homovanillic acid, 4-hydroxybenzoic acid, vanillic acid, 3-HPPA, trans-ferulic acid, caffeic acid, gallic acid, 3,4-dihydroxyphenyl acetic acid and urolithin B. Taken together, the neuroactive potential of bioactive gut microbe-derived metabolites of dietary polyphenols is promising but the mode of action of these metabolites need further in depth elucidation.

#### Phenolic Compounds

Ferulic acid (FA), a phenolic compound, has been the focus of a handful of studies researching the GBA and has been proposed a role in cognitive development and neuroprotection. Sources of ferulic acid are plants and seeds in the human diet, as well as from gut microbial biosynthesis from dietary cyanidin, catechin and epicatechin. Some beneficial properties of FA are protection from oxidative neurological damage by ROS scavenging, neural stem cell stimulation, and the direct inhibition of Aβ aggregation [8,212–214].

Dysfunctions in Pavlovian fear extinction learning are involved in anxiety and fearassociated neuropsychiatric disorders such as PTSD [215,216]. Recently, four bacterial metabolites, of which three were phenolic compounds (phenyl sulphate, pyrocatechol sulphate and 3-(3-sulfooxyphenyl)propanoic acid), have been shown to be significantly decreased in CSF, serum and faecal samples of GF or antibiotic treated mice showing defective fear extinction learning. Further investigations discovered alterations of gene expressions in the medial prefrontal cortex, immature-like microglia, as well as perturbed structural and functional changes in neurons involved in learning processes [168].

#### 4.2.3. Bacterial Amyloid Proteins

Amyloid proteins produced by bacteria have been an emerging subject of interest in the study of pathophysiology in PD [109]. It was reported that bacterial amyloid proteins, such as curli from *E. coli*, could induce the formation of human amyloid aggregates by cross-seeding in a prion-like fashion, as well as promote inflammatory processes by molecular mimicry [217]. The mechanism of cross-seeding was previously shown by curliinduced serum amyloid A amyloidosis in mice [218]. This was further supported by a hypothesis that the phenomenon of protein misfolding in neurodegenerative diseases might originate from the gut, possibly via bidirectional vagal fibres bypassing the circulatory system [108]. It is important to note that vagotomy has been associated with a lower risk for PD [219] and a delayed αSyn dissemination when nervous structures connecting the gut to the brain are severed [220]. Findings of αSyn aggregates in regions beyond the brain, such as the enteric nervous system and olfactory bulb, point to the phenomenon that the majority of PD patients experience gastrointestinal and olfactory dysfunctions long before their diagnosis [221]. Interestingly, hyposmia has been associated with cognitive decline, even in the context of AD [222]. A recent study observed that mice with αSyn overexpression developed PD-like pathological traits after inoculation with curli-producing *E. coli*. These mice displayed αSyn-aggregates in brain and gut tissues, as well as disturbed motor and intestinal functions [223]. Furthermore, prior research reported significantly increased amounts of αSyn aggregates in aged Fischer 344 rat brains and in *Caenorhabditis elegans* after oral inoculation with curli-producing *E. coli* in comparison with subjects colonized with identical strains lacking curli production. Moreover, exposure to curliproducing *E. coli* stimulated immune activity in rat brains, represented by enhanced neuroinflammatory markers, such as upregulated Toll-like receptor 2 (TLR2), interleukin 6, tissue necrosis factor, microgliosis and astrogliosis [224]. The similar immune response pathways to bacterial amyloids by pathogen-associated molecular pattern recognition, and the response to misfolded endogenous amyloids like αSyn and Aβ is intriguing. This

illustrates the possibility that gut bacterial amyloids could prime the immune system for a neuroinflammatory response to cerebral amyloid deposits, thus leading to enhanced neuroinflammation and degeneration [217].

Species reported for curli-production are from the family *Enterobacteriaceae*, including *E. coli*, *Salmonella typhimurium*, *Citrobacter spcc.*, *Cronobacter sakazakii* and *Proteus mirabilis.* This was recently deemed as conspicuous by researchers that observed enriched *Enterobacteriaceae spp.* in 31% of PD studies, as well as previous associations between *P. mirabilis* and a PD mouse model [109]. Other residents of the human gut were also reported to generate extracellular amyloids, namely *Streptococcus*, *Staphylococcus*, *Mycobacteria*, *Klebsiella* and *Bacillus spp*. [217]. Overall, these studies provide a hypothesis that bacterial amyloids could be key players in the pathogenesis of neurodegenerative diseases like PD and AD, with data pointing to their effects on amyloid aggregation and on enhanced immunoreactions in the CNS. This subject even motivated Friedland et al. [217] to propose the new term "MAPRANOSIS—the process of microbiota-associated proteopathy and neuroinflammation". However, some preclinical studies have implicated bacterial amyloids with increased clearance and decreased neuroinflammation by means of activating microglia through the receptor TREM2 [91]. Similarly, toll-like receptors 2 and 4 (TLRs) have shown contradicting effects on amyloid-related neurotoxicity and clearance through microglial activation [225]. It is difficult to draw definite conclusions regarding the effect of gut bacterial amyloids on AD and PD. Nevertheless, current data are strongly pointing to an existing connection.

Yang et al. [226] recently conducted the first study to observe implications of gut dysbiosis and faecal metabolic changes in mice with prions disease, thus providing the base for a new area of research in brain diseases with links to the gut microbiome. Of the previously mentioned metabolites, SCFAs and Trp were decreased, while Tyr increased in mice with prions disease. Furthermore, the microbiome of prion-infected mice harboured distinct compositions from HC, namely enriched Lactobacillaceae, Helicobacteraceae and decreased Prevotellaceae and Ruminococcaceae. Additionally, newly observed altered faecal metabolites consisted of various glycerophospholipids, three secondary bile acids and the toxic avermectin A2b. However, further research should investigate if these compounds are derived from gut bacterial metabolism and whether they are biomarkers or mediators of disease.


**Table 8.** Gut-residing bacteria found to correlate with the production of other metabolites.

TMA(O) *=* trimethylamine N oxide; TMA *=* trimethylamine; 3-HBA *=* 3-hydroxybenzoic acid; 3,4-diHBA *=* 3,4-dihydroxybenzoic acid; DHCA *=* dihydrocaffeic acid. References are represented by numbers in brackets.

#### **5. Discussion and Conclusions**

Our collected data highlight the fact that research into GBA and the precise role of bacterial metabolites as key contributors is still in its infancy. The majority of studies were conducted in preclinical animal or cell models and only a limited number of human studies are contributing to the current knowledge.

Only a handful of human studies of brain diseases reported faecal SCFAs levels, making it difficult to draw definitive conclusions. Our literature search captured five studies in ASD, two in affective disorders, two in MS, one in PD and none in AD as of this point in time. Nevertheless, an overall decrease in faecal SCFA levels in ASD, affective disorders, MS and PD is apparent (Table 1). Importantly, an increase in one SCFA may be levelled out by the decrease in another SCFA in the same study. AA and PA were reported to be increased in faecal samples [39,42] in two studies, while BA was increased in one [42]. Furthermore, one study observed no significant changes across all SCFAs in ASD [41]. Therefore, it remains to be determined by in-depth studies of the human intestinal metabolome, whether distinct patterns of SCFA levels are present and relevant in different brain diseases.

Based on preclinical findings on how SCFA might impact the brain, the overall assumption points to a beneficial role. The non-exhaustive list of reported effects includes improved gut barrier and BBB integrity [19–22] and an overall shift towards anti-inflammatory processes [13,28,29,74,75,91]. As HDACI, SCFAs can epigenetically modulate the maturation of brain cells [28,84,87], enhance gene expression for enzymes relevant in catecholamine production [23] and even shift the balance of the immune system towards anti-inflammatory Treg cells and away from pro-inflammatory Th1 and Th17 cells [13,81,83]. Inflammatory processes are known to be involved in MS and to take part in neurodegenerative diseases. As SCFA levels tend to be decreased in most neurodegenerative diseases (Table 1), we might assume that a perturbed gut microbiome might lead to impaired SCFA production of gut bacteria, which might then deplete the beneficial anti-inflammatory effects on the CNS.

Regarding non-SCFA bacterial metabolites, p-cresol has yielded the most human data across studies. It is also the only metabolite with measurements in faecal samples that correlated with brain disease, namely in patients with ASD. Taken all results together (Table 6), p-cresol is significantly increased in faecal, urinary and blood samples of ASD, as well in the CSF and serum of PD patients. CSF and plasma levels of p-cresol have moreover shown to be correlated with the severity of PD [163]. This, and preclinical studies in mice models implicating p-cresol with detrimental effects on the CNS [140,141], seem to be in line with elevated levels in ASD and PD. The notion of involvements of p-cresol in ASD is supported by the finding that 4EPS, a metabolite with structural similarities to p-cresol, induces ASD-behaviour in mice.

Alterations in Trp metabolism were observed in human studies in ASD, PD and MS. Significant assessments of Trp metabolites such as indole and indole derivatives are mainly from non-faecal samples and have shown an overall increase in ASD, a decrease in MS and inconsistent results in PD (Table 6). Nevertheless, it is important to mention that two PD studies showed indole and indole derivatives to be increased in CSF [109, 163], a compartment closely connected to brain tissue. Furthermore, metagenomic tests concluded that overall Trp metabolism pathways are enriched in the gut microbiome of ASD patient cohorts, and that indole pathways are enhanced in PD microbiomes [135]. These metagenomic findings and the several reports on metabolite levels in patients seem to indicate the presence of perturbed and enriched Trp metabolism in ASD and PD. Naturally, the lack of studies testing for faecal metabolites in humans, as well as the inconsistencies within results in PD stress the limitation to unequivocally determine distinct patterns by which our gut bacteria alter Trp metabolism in various brain diseases.

Several probable mechanisms by which Trp metabolites might exert their impact were identified. Preclinical studies have indicated that indole and its derivatives (indole, IS, IPA, indole-3-aldehyde, IAA) might actually be able to limit neuroinflammation by acting as agonists on the AHR [109,182] (Figure 3). Furthermore, a shift in Trp-metabolism away from 5-HT production might partly explain the worsening of disease in individuals who are responsive to treatment with selective serotonin-reuptake inhibitors (SSRIs) [146,147]. Thus, a decrease in Trp availability might be connected to brain diseases, which is further

exemplified by Trp-depletion experiments in depressed human cohorts and mice models of MS [146,147,182]. It might further be of interest that IPA was reported to ameliorate toxic ROS activity in the brain [135]. On the other hand, IS has induced oxidative stress in animal and in vitro studies [163,164], as well as displayed potential neurotoxic effects in mice through perturbed microglial and astrocyte function [165]. Moreover, an overproduction of indole was associated with anxiety and depression levels in rats [155]. Taken together, these results imply that alterations in the gut bacterial metabolism of AAA, whether it is Tyr-derived p-cresol or Trp-derived indole metabolites, could contribute to brain diseases.

The question as to whether gut bacterial amyloids like curli might contribute to or alleviate diseases with an accumulation and aggregation of misfolded proteins and neuroinflammation, cannot be unequivocally answered yet. Nevertheless, the current evidence on bacterial amyloids implies that there might be more to bacterial metabolites with connections to the brain beyond SCFAs and AAM.

The majority of evidence presented here is derived from preclinical studies, such as in vivo studies in transgenic animals, in germ-free animals or animals exposed to earlylife alterations of the gut microbiota including pathogens, probiotics, or antibiotics. The validity of drawing decisive conclusions for the human physiology from animal studies is therefore conceivably limited. Moreover, the information on taxonomical alterations and associations to metabolites and diseases is non-exhaustive since this study focused on bacterial metabolites. The objective was to assess bacterial metabolites independently as probable determinants of disease, thus shifting the focus away from their producers. The need for further work in deciphering the vast and intricate correlations between gut microbial communities, faecal metabolites and the presence of brain dysfunction, is definitely acknowledged and deemed as necessary.

Results regarding the potential protective or aggravating role of bacterial metabolite groups on brain diseases are still few and require additional confirmation. A final assessment of the importance of faecal metabolomic changes in brain diseases, however, is problematic as of now. The difficulties lie in the heterogeneity of disease manifestations and varying technological methods to assess varying sample sources. Furthermore, confounders like gender, genetics, dietary factors, medication and lifestyle, as well as subtypes of bacterial species, might contribute to insignificant or falsified results. Therefore, additional independent research applying methodical standardization is essential to ensure comparable and reproducible data. Determining not only taxonomical data, but also conducting functional analyses through metagenomic and metabolomic testing of faecal samples would crucially increase the robustness of discovered associations. This might further develop, confirm or refute today's initial findings on correlations between bacterial metabolites and brain diseases. Interests lie in the discovery and unravelment of GBA mechanisms, as well as the study of bacterial metabolites as promising key contributors to brain diseases. Successful findings of such might be crucial in identifying aetiological and pathophysiological processes, thus efficaciously supporting future research in novel treatments and the prevention strategies of brain diseases.

**Author Contributions:** Conceptualization, S.M.-S.T. and M.H.M.; methodology, S.M.-S.T. and M.H.M.; validation, M.H.M.; formal analysis, S.M.-S.T. and M.H.M.; investigation, S.M.-S.T.; resources, S.M.-S.T.; data curation, S.M.-S.T.; writing—original draft preparation, S.M.-S.T.; writing review and editing, S.M.-S.T. and M.H.M.; visualization, S.M.-S.T.; supervision, M.H.M.; project administration, M.H.M. and S.M.-S.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data sharing not applicable. No new data were created or analyzed in this study. Data sharing is not applicable to this article.

**Acknowledgments:** The authors thank David P. Wolfer for valuable comments.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Nutrients* Editorial Office E-mail: nutrients@mdpi.com www.mdpi.com/journal/nutrients

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-3803-7