Recent studies have discussed the fundamental connection between the gut microbiome and the behavioral symptoms associated with the Autism Spectrum Disorder (ASD). [1, 2, 3, 4, 5]

The gut microbiome has been characterized as an important mediator in the communication between the gut and the brain and as a key modulator in the peripheral immune response that alters the microglia and initiates the neuroinflammatory processes that have been linked to the behavioral symptoms (impairment of cognitive functions, emotional regulation, speech, etc.) traditionally associated with the Autism disorder.

The connection between the gut microbiome and Autism severity has been further established by a recent open-label clinical trial conducted by the Arizona State University (ASU) using Microbiota Transfer Therapy (MTT) consisting of the resettlement of gut microbiota obtained from carefully selected donors into trial subjects. Astonishingly, trial participants observed a near 50% reduction in Autism symptoms across the board and an 80% reduction of GI problems 2 years post-MTT.

Still, some researchers have observed that there isn’t “a single definitive Autism microbiome pattern” and that “more research is needed to better define the connection” between the gut microbiome and Autism severity.

Hence, building on the learnings discussed in Part 1 of this series, this article presents,

• An augmented (and more accurate) version of a predictive model linking Autism severity to gut bacteria counts found in the stool reports of 24 ASD-diagnosed children (ages 2 to 7) across multiple bacterial species
• A list of common interventions that can drastically impact bacteria counts that (if not taken into account) may significantly obscure the understanding of the gut-brain connection in clinical research and the identification of potential treatment opportunities that might lead to significant reductions in Autism symptoms, as shown in the MTT trial

Bacteria-altering Interventions

According to genetic reports, Autistic behaviors may be understood as the ultimate manifestation of a series of genetic “combinations of risk” that may trigger (or inhibit) key enzymatic pathways for the production and metabolism of certain neurotransmitters that are responsible for the regulation of anti-inflammatory processes in the body.

Common genetic combinations of risk may affect the activity of enzymes, genes and receptors that participate in the elimination of toxins and heavy metals and in the production and metabolism of neurotransmitters associated with the regulation of stress, emotional behavior, sensitivity to pain and with the release of hormones that could lead to behavioral disorders such as self-injury, anxiety, excitability, nervousness, changes in mood, sleep disorders, tics, spasms, and repetitive and/or involuntary movements.

Consequently, its is very common to find ASD-diagnosed individuals under very strict dietary, supplementation and detoxification interventions to support the immune system, the activity of enzymes, genes, receptors, neurotransmitters, and the elimination of toxins and heavy metals.

Although highly necessary at times, some of these regimens can have a substantial impact on the intestinal bacterial populations which might obscure the health state of the gut microbiome in connection to Autism severity, if such factors are not taken into account.

Here are some examples,

#1. Chelation, Antibiotics, Antifungals & Antiparasitics

The following chart shows the effect of a series of interventions on the gut bacteria counts of an ASD-diagnosed boy aged 2 to 5, from March 2016 to January 2019. Six comprehensive stool reports were produced by the same laboratory that used the same counting method in colony-forming units per gram (CFU/g) of stool throughout this period.

This chart shows the 7 species that were found to be most highly correlated with ATEC (see Part 1) as a function of the various interventions this boy received during this 3-year span.

The time period for each intervention is represented by the width of the blue boxes, i.e. the boy was reported to be on a GAPS diet between March 2016 and February 2017, on an antiparasitic drug around the summer of 2016, on an antibiotic in early 2018, on IV chelation between September 2017 and February 2018, and took various probiotics and prebiotics between January 2017 and January 2019.

The table below shows the boy’s CFU/g counts reported at each point in time with a Total roll-up across the 7 species under study,

CFU/g	Mar 2016	Jan 2017	Jul 2017	Feb 2018	Jul 2018	Jan 2019
Escherichia coli	5.00E+06	4.00E+08	2.00E+04	2.00E+04	2.00E+04	5.00E+06
Bifidobacterium	5.00E+09	4.00E+08	1.00E+09	4.00E+07	1.00E+08	3.00E+09
Bacterioides	6.00E+09	9.00E+09	1.00E+10	8.00E+08	4.00E+08	1.00E+10
Lactobacillus	4.00E+06	4.00E+05	1.00E+05	2.00E+04	2.00E+04	2.00E+04
Clostridium	2.00E+04	2.00E+04	2.00E+04	2.00E+04	2.00E+04	4.00E+05
Faecalibacterium prausnitzii	2.00E+08	9.00E+08	5.00E+08	2.00E+08	5.00E+08	5.00E+09
Akkermansia muciniphila	3.00E+02	4.00E+07	9.00E+07	3.00E+07	1.00E+07	1.00E+04
Total	1.12E+10	1.07E+10	1.16E+10	1.07E+09	1.01E+09	1.80E+10

The Total figures show that the use of antibiotics (Metronidazole) and the IV chelation protocol had an impact of over 10 billion colony-forming units per gram of stool, i.e. 1.06E+10.

That is, before the antibiotic/chelation protocol started (July 2017) there were 1.16E+10 CFU/g and by the end of the protocol (July 2018) there were 1.01E+09 CFU/g, which represents the eradication of 91.3% of the bacterial population there was in this boy’s microbiome, on these 7 species alone.

In spite of the drastic impact in Total CFU/g, no noticeable changes in ATEC were observed during this period indicating that changes in the microbiome may take time to be reflected in the person’s overall health and symptoms. The boy’s parents did report however a slight improvement in focus and speech when the bacterial populations where regained thanks to the use of pro/prebiotics several months after the antiobiotic/chelation protocol ended, in early 2019.

#2. Probiotics & Prebiotics

The use of probiotics and prebiotics may also significantly alter gut bacteria counts in stool profiles. It is suspected however that the observed changes registered in stool samples due to the use of probiotics may not be representative of the actual microbiotal diversity of the species settled in on the host’s microbiome.

Recent studies have shown that the relative benefits of industrialized probiotics may be significantly influenced by the gut gene expression profile of the recipient. Individuals may be generally characterized as “persisters”, i.e. those whose guts may be successfully colonized by off-the-shelf probiotics, or as “resisters”, i.e. those whose guts will expel the probiotics before they can settle in the gut lining.

Besides the gene expression profile, it has also been theorized that the person-specific resistance to probiotics might be also determined by the overall state of the gut microbiome and by the host’s immune system at the time of the intake, which may be directly correlated with the state of the gut lining of the host and the microbiome diversity.

#3. Food

If consumed regularly or early before stool samples are collected, certain types of food may also have probiotic effects which may temporarily increase bacterial counts in stool profiles,

• Yogurt, kefir and other dairy derivatives like buttermilk and fermented cheese may contain multiple strains of active or live cultures
• Honey and other bee derivatives such as pollen, propolis, royal jelly, et al. have powerful anti-microbial properties that may act as a prebiotic in the GI tract boosting the population of certain bacterial species
• Sauerkraut, tempeh, miso, kombucha, pickles and kimchi also have prebiotic fermenting prebiotic and probiotic properties that will act as phages to significantly increase certain bacteria species after consumption

Augmented Predictive Model: Gut Bacteria on Autism Severity

Building on the findings discussed in Part 1 of this series, the Author collected comprehensive stool profiles from parents located in the United States and Spain that voluntarily submitted stool profiles for a total of 17 ASD-diagnosed children (11 of which were part of the initial sample in Part 1) along with their ATEC scores as a measure of Autism severity at the time of the stool test.

All lab reports considered in the analysis were produced by Genova Diagnostics in the United States and by the Instituto de Microecología based in Madrid, Spain. Both labs reported colony-forming units per gram (CFU/g) of stool across multiple bacteria species, genus and phylum.

A couple parents submitted multiple stool reports for the same child (e.g. #1 above) for a total of 24 reports submitted. However, after careful examination, 11 reports were excluded as parents reported that their children were receiving one or more of the interventions listed in #1–3 at the time the stool sample was taken.

To understand the effect these interventions had in identifying potential treatment opportunities, comparative analyses were run on both sets of data, i.e. those containing the 13 reports that did not present any of the confounding interventions vs. all 24 cases.

The following table shows a side-by-side comparison of the bivariate correlations of the bacteria counts in each dataset on ATEC,

Species/Genus	Correlations (ρ) Including All Reports	Correlations (ρ) Excluding Confounding Interventions
Clostridium	-0.32	-0.52
Escherichia coli	-0.22	-0.48
Faecalibacterium prausnitzii	-0.00	-0.48
Bifidobacterium	-0.33	-0.48
Lactobacillus	-0.06	-0.45
Akkermansia muciniphila	-0.12	-0.40
Bacterioides	0.05	0.01

Both the magnitude and the sign of the correlation coefficients are meaningful. Negative correlations indicate an inverse proportional relation between gut bacteria counts and ATEC, i.e. the lower the bacteria counts the higher the ATEC and the more severe the condition.

Correlation coefficients range between −1 and +1 and represent the extent in which one variable may be explained (or inversely explained) by the presence of another. Correlations around 0 mean that the presence of one variable is not directly related to the presence of another.

It is clear from the bivariate correlations that the effect of the interventions described on bacteria counts was very significant. While 6 of the 7 species were highly correlated with ATEC (ρ < -0.40) when the interventions were considered, 4 out of 7 had non-significant correlations when the impact of those interventions was ignored.

Although the magnitude of the correlations is indicative of the strength of the relationships, a multivariate model allows for the prediction of ATEC using all bacteria counts across all species simultaneously. Hence (consistent with Part 1), ordinary least squares (OLS) regression models were used to evaluate the predictive power of both sets of data on ATEC,

Remarkably, the predictive power of the OLS model that excluded the confounding interventions was of 85.6% (a 3-percent point improvement compared to the previous iteration) whereas the one that included all cases was significantly lower, at only 32.6%

This important disconnect in predictive power between the 2 models highlights the importance of accounting for any interventions that might affect the microbiome in clinical research and when evaluating treatment opportunities for Autism patients in clinical practice.

Of particular interest also is the shift in the order bacterial species were entered into the OLS model compared to the iteration presented in Part 1.

Given the “forward stepwise” nature of the OLS approach used, the first variable entered represents the species that happened to be most highly correlated with ATEC, whereas the remaining species are added on the basis on their statistical significance in explaining ATEC in a way not yet explained by the species already added into the model.

Important changes in the order variables enter predictive algorithms from one iteration to another is expected in studies with limited samples. In this case however, the shift may also signal the inherent interdependencies that microbiotic ecosystems are subject to as some species, e.g. Akkermansia muciniphila, will feed from the richness of the mucus layer (or mucin) afforded by certain species, e.g. Lactobacillus, and by other species whose dead bacteria may be used as prebiotic phages to feed co-dependent bacterial organisms.

Hence, microbial interdependencies will be expected to have an overlapping effect in linear models like the OLS as all species participate collectively in supporting the immune response and the neuroinflammatory processes that are ultimately manifested in the Autistic behaviors.

Beyond its predictive power, the appropriateness of the OLS method is also given by a series of diagnostic elements to examine the underlying assumptions of normality and homoscedasticity. A visual examination at these assumptions confirms that the regression technique used (with all confounding factors removed) is a valid and appropriate method,

Although generally regarded as a highly subjective metric, the ATEC scale is a useful tool to gauge the overall severity of the Autism disorder. The following table shows the parental assessments vs. the predicted ATEC values as per the OLS model,

id	ATEC	Predicted
1	65	58
2	40	55
3	45	46
4	20	20
5	85	90
6	42	42
7	44	47
8	95	84
9	62	75
12	50	47
14	80	68
15	50	49
17	45	42

Remarkably, all 13 cases included in the analysis were predicted within a +/- 15-point range and 10 of them were predicted within a +/- 7-point range of the ATEC scores provided by parents, which may be in part attributed to the subjectivity of the scale.

Also, it should be noted that the lab reports used rounded all CFU/g figures to the nearest thousand, million, billion, etc. Given that the average figure across all species and subjects was 1 billion (1E+09) CFU/g, it is clear that some “noise” in the prediction rates could be attributed to the numerical rounding of the (very large) figures used to report bacterial counts.

Interestingly, the OLS model was also able to accurately discriminate on neurotypical (NT) controls. Bacteria counts on the 7 species under study where collected from a couple of NT controls and the model (correctly) predicted negative ATEC values, i.e. signaling that those 2 individuals were in fact not on the Autism spectrum.

NT id	ATEC	Predicted
1	N/A	-760
2	N/A	-30

Conclusions

Although still applied to a limited sample, the consistently high predictive power of the OLS algorithm continues to support the findings of the MTT-ASU phase-I clinical research. Higher bacteria counts and more diverse microbiomes were in fact strongly associated with lower ATEC scores and vice versa, lower counts and less diverse microbiomes were strongly associated with higher ATEC scores.

Consistent with the MTT-ASU report, Bifidobacterium and Bacteroides-Prevotella were found to be particularly important in determining Autism severity. However, the species Clostridium, Faecalibacterium prausnitzii, Lactobacillus and Akkermansia muciniphila were also found to be highly correlated with ATEC in bivariate comparisons and when considering all species simultaneously.

The impact of common dietary, supplementation and detoxification interventions can have a substantial impact in gut bacteria populations which may significantly confound clinical research and potential treatment opportunities which may lead parents and physicians astray, if not accounted for.

Strategies for Autism treatment must consider the role of the microbiome in supporting the immune system, reduce toxicity and alleviate neuroinflammatory processes traditionally responsible for the Autism symptoms and behaviors.

Disclosure

The Author has no financial interest in neither Genova Diagnostics or Instituto de Microecología labs or in any other lab performing stool testing, is not involved in any way with the MTT-ASU research team, and worked pro bono in the collection of the data, analysis, and in the writing of this article.

Acknowledgements

A very big “Thank you!” to all the parents that provided the comprehensive stool reports, ATEC scores and additional details on the dietary, supplementation and detoxification interventions for their children. Special thanks to Natalia Marmol, Mónica Cuenca and Jenni McKay who were instrumental in the data collection efforts to make this research possible.