Assessment of haptoglobin alleles in autism spectrum disorders

Gene-environment interactions, by means of abnormal macromolecular intestinal adsorption, is one of the possible causes of autism spectrum disorders (ASD) predominantly in patients with gastrointestinal disorders. Pre-haptoglobin-2 (zonulin), encoded by the Haptoglobin (HP) allele-2 gene, enhances the intestinal permeability by modulation of intercellular tight junctions. The two alleles of HP, HP1 and HP2, differ for 2 extra exons in HP2 that result in exon duplication undetectable by classic genome-wide association studies. To evaluate the role of HP2 in ASD pathogenesis and to set up a method to discriminate HP alleles, Italian subjects with ASD (n = 398) and healthy controls (n = 379) were genotyped by PCR analysis; subsequently, the PCR results were integrated with microarray genotypes (Illumina Human Omni 1S-8), obtained using a subset from the same subjects, and then we developed a computational method to predict HP alleles. On the contrary to our expectations, there was no association between HP2 and ASD (P > 0.05), and there was no significant allele association in subjects with ASD with or without gastrointestinal disorders (P > 0.05). With the aid of bioinformatics analysis, from a window frame of ~2 Mb containing 314 SNPs, we obtain imputation accuracy (r2) between 0.4 and 0.9 (median 0.7) and correct predictions were between 70% and 100% (median 90%). The conclusions endorse that enhanced intestinal permeability in subjects with ASD should not be imputed to HP2 but to other members of the zonulin family and/or to environmental factors.

. HP alleles and protein structures. (a) HP1 and HP2 allele structure; HP1 allele contains 5 exons, while HP2 is made up of 7 exons: exons 1 to 4 correspond to HP1, exons 5 and 6 is the duplication of exon 3 and 4, and exon 7 corresponds to exon 5 in HP1. (b) HP isoform and protein structure. HP1 allele encode for alpha-1 and beta chain (in blue and orange, respectively), while HP2 allele for alpha-2 and beta chain (green and orange respectively). The quaternary structures of HP genotypes are also illustrated (modified from 17 ). (c) HP gene encode for signal peptide, from amino-acid residues 1-18, and mature protein sequence from 19 to 347 or 406, depending on the allele of origin: HP1 allele encoding for alpha-1 chain (residues 19-101) and beta chain (d), HP2 allele for alpha-2 chain (residues 19-160) and beta chain (e). Alpha-1 chain lacks of residues 29-87 of alpha-2 chain, and beta chain is common to HP1 and HP2 alleles.
www.nature.com/scientificreports www.nature.com/scientificreports/ the two alleles, it is very difficult to recognize them by means of classic genome-wide association studies (GWAS), so most of the genotyping studies are limited to small size populations analyzed by PCR or quantitative RT-PCR.
Due to the protective role of HP1, HP1-1 genotype has a significant association to longevity 22 , on the other hand, HP2-2 has a higher susceptibility to immune-mediated diseases, such as celiac disease, Rheumatoid Arthritis, Type 1 diabetes, Systemic Lupus Erythematosus [23][24][25][26] and cancer.
In 2000, Wang and collaborators reported that an 8 amino-acid sequence in N-terminal of human pre-HP2 (residues 8-15: G G V L V Q P G) shared a common motif with the active fragment of Zot (residues 291-298-: G R L C V Q D G) from which originated the alternative name "zonulin". The common amino-acid motif is: non-polar (G), variable, non-polar, variable, non-polar (V), polar (Q), variable, non-polar (G) 13 . In 2018, Scheffler and collaborators identified another member of the zonulin family, properdin, a mannose-associated serine protease 27 .
Zonulin induces the disassembling of intercellular tight junctions in many epithelial and endothelial barriers, including the "leaky gut" and enhancing the intestinal trafficking of microorganisms and macromolecules that, in sequence, challenge the immune response involving different tissues and organs in genetically susceptible subjects 12,26,28 .
Of interest, a significant high level of zonulin (pre-HP2) in a small cohort of subjects with ASD vs healthy controls was found 29 , as well as an association between HP2 allele and autism with GID comorbidity 30 .
In this study, we genotyped a larger cohort of patients and controls to test the hypothesis that HP2 allele is associated with ASD development at least with GID in autism. We assessed the HP alleles in a batch of Italian subjects with ASD (n = 398) and healthy controls (n = 380) by PCR analysis. Because of the internal exon duplication of HP, it is very difficult to distinguish the two different alleles by means of standard bioinformatics analysis, and PCR techniques limit the size of tested subjects. We integrated PCR and microarray genotyping data, previously obtained on a subgroup of patients and controls, to impute and discriminate HP alleles by GWAS. Thus, we provide a bioinformatic reference useful for further HP prediction on publicly available GWAS data.  31 and Autism Diagnostic Interview -Revised (ADI-R) 32 . Subjects with genetic syndromes, epilepsy and neuroradiological confirmed disorders were excluded. Clinical data about the presence/absence of GIDs were evaluated in a sub-group of 157 subjects with ASD [135 males (86%) and 22 females (14%)].

controls. IRCCS
After denaturation for 30 seconds at 98 °C, thermo-cycling profile consisted of denaturation for 12 seconds at 98 °C, annealing for 10 seconds at 65.4 °C and extension for 1 minute and 40 seconds at 72 °C repeated for 35 cycles when using primers NewA and NewB, while denaturation for 2 minutes at 94 °C, thermo-cycling profile for 1 minutes at 94 °C, 40 seconds at 68 °C and 50 seconds at 72 °C repeated for 32 cycles, and extension for 2 minutes at 72 °C for primers NewC and NewD. PCR products obtained by NewA-B and NewC-D couples of primers were run on 1% and 2% agarose gel (E-Gel agarose gels, Invitrogen ™ , Thermo Fisher Scientific) respectively. In HP1-1 we detected the NewA-B PCR product of 1775 bp length, but not NewC-D PCR product; in HP2-2 we found the NewA-B PCR product of 3487 bp and NewC-D 360 bp PCR product; in HP1-2 subjects we identified both 3487 bp and 1775 bp amplicons, and the 360 bp PCR product (Fig. 2).

Sanger sequencing.
To check the PCR products, two PCR products for each type of HP genotype obtained by NewA and -B amplification (HP1-1, HP1-2, HP2-2) as well as those obtained by NewC and -D, were purified (CleanSweep PCR Purification Reagent, Applied Biosystems, Thermo Fisher Scientific, Wilmington, DE USA) and sequenced by Sanger method using NewC and NewD respectively for forward and reverse sequencing. Results were compared to HP1 and HP2 sequences (NCBI Reference Sequence: NG_012651.1).

Data analysis. The association between HP alleles and ASD was calculated considering subjects with ASD
vs all the analyzed controls and subjects with ASD without GID vs subjects with ASD with GID. Chi-square with Yates correction test was applied to evaluate statistical significance. We sub-classified the controls in two groups referring to the enrollment screening test: ASDs negative controls (n = 188), referred as non-ASD (NASD) controls, and those negative for ASD and DSM-IV Axis I disorders (n = 191), referred as super-controls. The HP allelic distribution was evaluated in all subjects with ASD and controls and in all sub-groups. Possible effect of sex on HP allele distribution was investigated in patients with ASD and in healthy subjects, considering the sub-classification of controls.
Due to the small size of controls enrolled in this study we implemented our data by a meta-analysis including HP genotyping of healthy Italians 22,36 , healthy Caucasians 35 (Table 2).

Microarray. Microarray analysis was performed on a subset of samples (n = 318) collected by IRCCS Eugenio
Medea -La Nostra Famiglia using Human Omni 1S-8 v 1.0 Illumina chip and iScan Illumina (Illumina, San Diego, CA, USA) according to the manufacturer's protocols, and a selection of 230 subjects were genotyped by microarray and PCR.
Among controls collected by the University of Brescia, 177 subjects (87 females and 90 males) were genotyped by Affymetrix Human Mapping GeneChip 6.0 array (Affymetrix, Thermo Fisher Scientific, Wilmington, DE USA) according to the manufacturer's protocols and applying quality control procedures described by Sacchetti and colleagues 37 , and 103 of them were genotyped for HP gene by PCR.
Bioinformatics. Genetic data for 1,185,076 SNPs in 230 subjects were collected from "Top" alleles of the Illumina's Genome Studio Final Report, generated using HumanOmni1S platform and GRCh37.p13 genome assembly. Standard quality control procedures were used to exclude individuals with discordant sex and call rates www.nature.com/scientificreports www.nature.com/scientificreports/ below 98% and filter out SNPs with MAF <1%, Hardy Weinberg p < 1 × 10 −4 and a call rate <99%. A total of 696,849 markers and 293 subjects fulfill quality control requirements. The analysis of genetic similarity among individuals revealed the presence of several outliers. Only individuals belonging to the most populated cluster were considered for further analyses. The quality control procedures repeated on this subgroup determined the exclusion of some additional markers. A total of 690'215 markers and 282 individuals were retained.
Genotypes detected by microarray and HP genotypes detected by PCR were phased using Beagle 38 against the haplotypes provided by the 1000 Genomes project (Phase 3). HP alleles were imputed using Beagle v4.1 39 . A total of 100 cross-validation trials were run, randomly assigning each time 90% of the subjects to the "reference" group and the remaining 10% to the "test" group.

Results
Hp allele distribution in patients and controls. As for PCR HP genotyping the length of NewA-B PCR products reflected the expected lengths (HP1 = 1775 bp; HP2 = 3487 bp) and included the duplication breakpoint sequence only in NewC-D products (360 bp) and in NewA-B products corresponding to HP2 allele (Fig. 2). . PCR products obtained with primers pairs NewC/NewD are 360 bp long in HP2 allele (b), while in HP1 no amplicon is given (c). Agarose gels, 1% (d) and 2% (e), are used to detect the three different genotype for both primer pairs. In the case of NewA and NewB in HP1-1 subjects only 1775 bp amplicon is present, in HP1-2 both amplicons are present and in HP2-2 only 3487 bp amplicon is detected (d). For NewC/NewD primer pairs 360 bp amplicons is detected only in HP1-2 and HP2-2 genotypes (e). Gels images (d,e) are cropped from images of full-length gels ( Supplementary Fig. S1).
The HP1 increased in subjects with ASD when compared to the total number of controls: Chi-square with Yates correction test was calculated and a significant increase of HP1 allele frequency was found in subjects with ASD compared to total controls (NASD controls+super controls) (P < 0.05) and super-controls (P < 0,005). Consequently, the association between HP1 and ASD disappeared (P > 0.05) excluding super-controls and adding to NASD controls, Italian and Caucasian healthy subjects 22,35,36 (Table 2).
The sex ratio differs from the subjects with ASD and controls, and this could influence the HP allelic distribution. For this reason, we also investigated HP genotype related to sex. HP allelic distribution did not show statistically significant imbalanced sex ratio in patients with ASD (P > 0.1), in the total number of controls (P > 0.5), and in the sub-groups NASD controls and super controls (P > 0.1) (Supplementary Table S1). To know more about the consequence of the sex imbalance within controls and subjects with ASD we also re-analyzed data from the study of Brackenridge 40 that found a significant sex imbalance in HP1-2 distribution within Australian population. HP allelic distribution was calculated for healthy Australian population (Supplementary Table S1) and we did not find significant difference in HP distribution between males and females by Fisher test (P > 0.5).
Hp alleles imputation from Snp haplotypes. To assess whether HP alleles can be predicted by the haplotype of surrounding SNPs we performed a series of cross-validation trials, in which we split the autistic population genotyped with the Illumina array into two groups ("reference" and "test") panels, and imputed (Beagle v4.1) the HP alleles of test subjects, using the multi-SNP haplotypes surrounding HP and HP alleles detected by PCR in subjects assigned to the reference group. We considered the SNPs occurring up to 2MB around HP gene and found the highest accuracy for HP2-2 (0.97) using 79 SNPs (0.5 Mb window). Among HP genotypes, we found that genotype 1-1 was the most difficult to predict ( Table 3).
The same approach was applied to healthy subjects, genotyped by the Affymetrix platform (Table 4). We observed a slightly lower accuracy (e.g. 0.89 using 76 SNPs; 0.5 Mb window), but its trends relative to the number of surrounding SNPs considered and to HP genotypes were similar to those observed in autistic subjects.

Discussion
Wang and colleagues first found, in human intestinal tissue, a protein that they called zonulin, sharing a common amino-acid motif with Zot, the Vibrio cholera toxin acting on the intestinal wall disassembles tight junctions with consequent intestinal permeability 13 . A following study identified human zonulin, as pre-HP2 12 . Since an involvement of zonulin has already been demonstrated in many immune mediated diseases with GID 41 , and many subjects with ASD suffer from GID and "leaky gut", a possible role of zonulin in ASD pathogenesis was assumed 29 . Nevertheless, because of its structure containing a duplication of 2 exons, HP gene is difficult to genotype by classic genome-wide studies and HP available data are limited to a very small amount of cases.
To understand the role of zonulin in ASD etiology and to provide a bioinformatics method to genotype HP using or re-using newly generated or archival data, respectively, we i) analyzed, by PCR, the HP allele distribution  Table 2. HP allelic distribution in subjects with ASD and controls. Data on neurotypical Italian population were from the total of controls enrolled by Bottini and collaborators and Napolioni and co-worker 22,36 . Genotyping data on Caucasian healthy subjects were from Koch and collaborators 35 , and HP allele frequency of individuals sampled by the 1000 Genomes Project was calculated in the study of Boettger and co-worker 58 . HP2 allele is highly represented in subjects with ASD and controls. HP1 allele increases significantly in patients when compared to total and non-ASDs controls enrolled in this study and to controls from 1000 Genomes projects. In contrast, HP1 allele did not increase significantly when compared to the other controls. P value was calculated with Chi-square with Yates correction test (*P < 0.05; **P < 0.005).
www.nature.com/scientificreports www.nature.com/scientificreports/ in a cohort of Italian subjects with ASD and healthy controls; ii) integrated PCR and microarray data of a subgroup of cases, to impute HP alleles from flanking SNP haplotypes and discriminate the two HP alleles.
Supposing that pre-HP2 corresponds to zonulin, we expected a very significant increase of HP2 allele frequency in subjects with ASD, or, at least, in cases suffering from GID. What we found is that HP2 allele prevailed in both subjects with ASD and controls. However, on the contrary to what we expected, there was a decrease in HP2 frequency when compared to the total number of controls (P < 0.05) in the observed subjects with ASD. Interestingly, this was due to a relative increase of HP2 in the super-controls rather than a decrease of HP2 in subjects with ASD. Indeed, comparing HP allelic distribution among different Italian control cohorts we found super-controls show a significant increase of HP2: P < 0.05 when compared to NASD and P < 0.005 when compared to those of previous study of Bottini and collaborators and Napolioni and co-worker 22,36 . Consequently, the association between HP2 and ASD disappears when the super-controls are excluded from the data analysis and decreases when comparing the ASDs only with the super-controls (Table 2).
NASD controls were unrelated ASD negative while super-controls belong to a group of a "selected healthy population", resulting negative when screened for any mental condition (DSM-IV Axis I disorders). In the group of super-controls HP alleles frequency was significantly different from that of Italian and Caucasian healthy subjects 22,35,36 (P = 0.0020 and 0.0006, respectively), while allelic distribution of NASD controls was similar to that reported in literature. This suggests that super-controls are a niche within unrelated non-affected population, therefore it could not be representative for the NASD population. Furthermore, HP genotypes distribution within worldwide population differs upon ethnicity and geographical area 42 . For these reasons, HP genotype data from previous studies on Italian population 22,36 were included among NASD controls. No significant differences of allelic distribution were found between ASDs and NASD controls + neurotypical Italian populations. One limitation of this study is the sex-ratio imbalance between controls and subjects with ASD, with an equal distribution of males and females in controls and a prevalence of males in patients. Literature on sex distribution of HP allele within healthy population showed no significant difference in sex-based HP allele distribution 40,43 . Indeed, the study of Zhao and colleagues declared no significant sex-based difference in the allelic distribution of HP; then again, the re-analysis of the data reported by Brackenridge showing that HP1-2 has different distribution between Australian males and females, revealed no significant differences in HP allelic distribution by Fisher test (P = 0.9066) (Supplementary Table S1). To better elucidate this important imbalance in our groups, we investigated the sex-based HP allelic distribution within subjects with ASD and controls to solve this bias. No differences were found between HP alleles in males and females in patients and total control, neither, separately, in NASD controls or in super controls. Taking together these findings, we conclude that HP allelic distribution is not sex dependent and that in these investigations the HP allele distribution is not related to sex imbalance between controls and patients with ASD.
A different age distribution between controls (adults) and patients (children) also exists. However, two previous study showed no significant association between age and HP genotype in healthy subjects 40,43 .
Moreover, no allelic association was found between subjects with ASD and patients suffering from GID. This indicates that the HP genotype does not represent a risk factor for ASDs pathogenesis, and that HP does not  Table 3. Bioinformatics imputation analysis, Illumina platform. Results of HP alleles imputation from SNP haplotypes genotyped using an Illumina array. Best prediction accuracy results are achieved in increments of 0,5 windows (in bold).

Number of markers
Accuracy of prediction  Table 4. Bioinformatics imputation analysis, Affymetrix platform. Results of HP alleles imputation from SNP haplotypes genotyped using an Affymetrix platform. Best prediction accuracy results are achieved in increments of 0,5 windows (in bold).
www.nature.com/scientificreports www.nature.com/scientificreports/ have a key role in the onset of GID in subjects with ASD. Furthermore, the significant decrease of HP1 allele in super-controls should be further investigated.
Our results do not reflect those of two previous studies that analyzed and found a relationship between zonulin and ASD. Indeed, Esnafoglu and collaborators 29 observed a statistically significant increase of serum zonulin in a group of 32 subjects with ASD compared with 33 healthy controls, by ELISA (Elabscience). Then again, HP genotype was determined in a cohort of 46 subjects with ASD (20 of which with GIDs and 26 without GIDs) and 41 controls (6 of which with GIDs and 35 without GIDs) by plasma immunoblot. This study also found an association between HP2-2 and ASD with GIDs when compared with neurotypical developing children (P < 0,01), and on the other hand found no association comparing ASD with GID with ASD 30 . However, these results were obtained from a very small cohort.
Considering the unexpected results of HP allele frequency and ASD, we have made some considerations based on literature and the zonulin ELISA commercial kit. Zonulin is often and wrongly considered the alternative name of HP. Indeed, many authoritative databases, including UniProtKB [(P00738 (HPT_HUMAN)], Nextprot (NX_P00738) and NCBI (Gene ID 3234), report both HP and zonulin among "names", while only pre-HP2 belongs to the zonulin family. This has induced some authors to consider zonulin and HP as the same protein. Furthermore, the zonulin ELISA commercial kits and antibodies also reported indifferently HP or zonulin [i.e. Human Zonulin ELISA Kit, Elabscience, (Wuhan, Hubei Province, China) and IDK ® Zonulin ELISA (Immundiagnostik AG, Germany)], and only after the publication of the study of Scheffler and colleagues, the datasheets were corrected reporting zonulin or HP as target protein.
Considering zonulin a family of proteins instead of a single one 27 modifies the interpretation of some previous literature. Indeed, zonulin quantification using commercial Human Zonulin ELISA kits was widely reported in different pathological conditions and the conclusions drawn on the basis of HP-zonulin identity. For instance, Esnafoglu and colleagues used the Human Zonulin ELISA Kit, Elabscience 29 to quantify zonulin in the serum of subjects with ASD. The capture antibody of this kit recognizes a sequence within residues 118-281 of pre-HP2/ zonulin (uniprot ID: P00738) and the detection antibody within residues 104-346 (as referred by technical support, but not reported in the datasheet https://www.elabscience.com/PDF/Cate61/E-EL-H5560-Elabscience.pdf). These sequences may include both alpha (not specified if alpha-1 or alpha-2) and beta chain of HP, and is not specific for the signal peptide characterizing pre-HP2 (Fig. 3). Then again, the capture antibody of the IDK ® Zonulin ELISA, recognizes a portion of zonulin previously reported by Wang and colleagues 13 , (datasheet http://www. immundiagnostik.com/fileadmin/pdf/Zonulin_K5601.pdf). Since this portion is not included in HP sequence, the antibody may recognize another zonulin family member, as suggested by Scheffler and colleagues 27 . Indeed, these authors, which integrated antibody capture experiments, mass spectrometry, and Western blot analysis, demonstrated that the IDK ® Zonulin ELISA (based on a polyclonal antibody anti-zonulin) mainly recognize properdin, a potential second member of the zonulin family 27 . Interestingly, properdin (P27918 https://www. uniprot.org/uniprot/P27918) maps on Xp11.23 making male subjects more susceptible to its allelic or variant effect. Furthermore, this chromosome region includes three other genes related to autism: the calcium channel, voltage-dependent, alpha 1 F (CACNA1F) associated with syndromic autism and schizophrenia 44 , the phosphatase 1, regulatory (inhibitor) subunit 3 F (PPP1R3F) from which rare mutations have been found in autism 45 and histone deacetylase 6 (HDAC6) from which a partial skipping of exon 3 was found in a subject with ASD 46 .
Then again, gut dysbiosis has been described in subjects with ASD 47-50 and gut microbial imbalance, that increases inflammation and microbial toxic metabolites production, is the strongest stimulus for the activation of the zonulin pathway with consequent intestinal permeability and trafficking of macromolecules through intestinal wall 51 . Moreover, gut microbes strongly participate in "microbiota-gut-brain-axis" (the microbiota-CNS cross-talk) 52 modulating inflammatory cytokines, neurotransmitters production and epigenetic factors such as RNA interference, DNA methylation and histone modification. So, microbiota alteration can produce negative effects on brain function via "gut-brain axis" dysregulation and, indeed, dysbiosis has been reported in many psychiatric conditions 53 including ASD [47][48][49][50] . Recently, the role of mycotoxins has also been proposed in ASD pathogenesis 54,55 . Mycotoxins, worldwide contaminants of food with toxicological effects, induce intestinal permeability and inflammation and interact with gut microbiota resulting in impairment of gut health 56 . www.nature.com/scientificreports www.nature.com/scientificreports/ It is well known that the HP CNV is not in strong linkage disequilibrium with any individual SNP and therefore it is not successfully genotyped by SNP-array technologies 57 . This is likely due to the fact that HP1 allele arose from recurrent deletions in HP2 allele 58 . In addition, HP1F and the left copy of HP2FS share a 300 bp sequence identical to a segment of Haptoglobin-Related Protein (HPR) gene. This HPR sequence that contains many derived variants was probably transferred into HP gene through a paralogous conversion event making HP genotyping unpredictable by classic GWAS 58 . Thus, PCR or Real Time-PCR is generally performed for HP genotyping on very small cohorts.
Although HP1 allele arose many times, it has been shown that it is possible to impute HP alleles from SNP haplotypes with a high level of accuracy 58 . Indeed, alleles that are old and common today segregate on characteristic SNP haplotypes. In this paper, we assessed the performance of HP alleles imputation from SNP haplotypes in our cohort. Overall, we observed a high accuracy in the prediction of HP genotypes in both autistic and healthy subjects screened by Illumina and Affymetrix platforms. In both groups, the prediction of HP2-2 genotype is more successful (accuracy is 0,97 in both platforms), on the other hand HP1-1 genotype is lower (0,80 and 0,59, respectively).
Beagle prediction based on the Illumina microarray platform gives acceptable results and should be useful to genotype HP for a large scale of subjects. Due to the lower accuracy of HP1-1 prediction, even if it represents a good accomplishment, a PCR genotyping to confirm HP1-1 prediction should be useful.
It is important to highlight that the intention of our study was to verify the distribution of HP2 allele in subjects with ASD and to relate this with GID comorbidity, also driven by the consideration of the increased levels of pre-HP2/zonulin in subject with ASD, measured by ELISA 29 . The unexpected results addressed the question on the possible dissimilarity between zonulin and HP2. During this investigation Scheffler and colleagues 27 and Ajamian and colleagues 59 preceded thus we confirm their statements.

conclusion
Zonulin is a family of structurally and functionally associated proteins including pre-HP2 and properdin. On the contrary to our expectations, no correlation between HP alleles and Italian ASD patients or between subjects with ASD and those patients suffering from GID was found. These results further support a recent study of Schleffer and colleagues 27 . They demonstrated that, within zonulin family members, properdin protein, rather than haptoglobin, is similar to Zot and possibly is involved in intestinal permeability. Interestingly, properdin maps on Xp11.23 show that male subjects are more prone to its effects. Further efforts should be dedicated to genotype and/or to sequencing this gene in subjects with ASD.
However, additional investigations with a wider number of cases and controls are necessarily required to confirm these results. For this purpose, the study proposes a bioinformatics method to predict HP allele distribution starting from GWAS data.
Moreover, many issues concerning HP and zonulin definition and detection must still be thoroughly studied. In conjunction with new genetic and/or environmental and predisposing factors which may lead to or provoke leaky gut in autistic subjects.