The study sample comprised seven individuals residing in a single living unit at the ASP Charitas residential facility in Modena, Italy. All participants met the criteria for ASD requiring “very substantial support” (Level 3), as defined by the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision (DSM-5-TR). Diagnostic confirmation was based on comprehensive clinical evaluations and supported by scores from the Childhood Autism Rating Scale, Second Edition – Standard Version (CARS-2-ST), administered by licensed clinicians with expertise in neurodevelopmental disorders. Participants were eligible for inclusion if they had a confirmed diagnosis of Level 3 ASD, presented with significant behavioral or psychiatric comorbidities that rendered family-based care unfeasible, and had resided at the facility for a minimum of three years to ensure environmental stability during the intervention. Exclusion criteria included the presence of unstable medical conditions, known food allergies that would interfere with adherence to the dietary protocol, or recent changes in psychotropic medications (within 60 days prior to study initiation). Detailed demographic and clinical information—including age, sex, and duration of residency—is provided in the Results section.

In our study, we selected specific biological indices to assess potential neurobiological dysfunctions in individuals with ASD. These indices and their rationales, supported by current literature, are as follows:

To assess the lipidomic status of participants, we analyzed the fatty acid composition of cell membrane phospholipids, with a specific focus on mature red blood cells (RBCs) (47). Due to their approximately 120-day lifespan and incorporation of fatty acids into membrane phospholipids during erythropoiesis (48), RBCs provide a stable and integrative measure of fatty acid availability over time, allowing for evaluation of dietary fatty acid intake and endogenous membrane remodeling processes associated with cellular growth and turnover (40–42). The membrane lipidomic analysis of RBCs was performed using an analytical protocol certified to comply with ISO 17025 requirements, ensuring the reliability and repeatability of the results. In particular, the analytical process involves the transformation of membrane phospholipids isolated from mature red blood cells into fatty acid methyl esters (FAME, then analysed by gas chromatography), using a mild alkaline procedure that is very respectful of the PUFA residues, particularly those of plasmalogens, known to be sensitive to acidic and oxidative conditions (49). The fatty acid composition of mature RBC membrane phospholipids was obtained from blood samples (about 2 mL) collected in vacutainer tubes containing ethylenediaminetetraacetic acid (EDTA). Samples were shipped to the ISO 17025-certified Lipidomic Laboratory (LAB #01622, Lipinutragen srl, Bologna) at a controlled temperature and, upon arrival, underwent quality control to verify the absence of hemolysis. During blood processing, before lipid extraction and the conversion to fatty acid methyl esters (FAMEs), the automated protocol included selecting mature RBCs, as previously reported (50). Briefly, the whole blood in EDTA was centrifuged (4000 revolutions per minute (rpm) for 5 minutes at 4°C), and the mature cell fraction was isolated based on the higher density of aged cells (50), with the process controlled using a cell counter (Scepter 2.0 with Scepter™ Software Pro, EMD Millipore, Darmstadt, Germany). All subsequent steps were automated, including cell lysis, isolation of membrane pellets, phospholipid extraction from pellets using the Bligh and Dyer method (51), transesterification to FAMEs by treatment with a potassium hydroxide (KOH)/methyl alcohol (MeOH) solution (0.5 mol/L) for 10 minutes at room temperature, and extraction using hexane (2 mL). FAMEs were analysed via capillary column gas chromatography (GC). GC analysis was performed on the Agilent 6850 Network GC System (Agilent, USA), equipped with a fused silica capillary column Agilent DB23 (60 m × 0.25 mm × 0.25 μm) and a flame ionisation detector. Achieving optimal separation of all fatty acids and their geometrical and positional isomers, identification, and calibration were performed by comparison to commercially available standards and a library of trans isomers of MUFAs and PUFAs. The quantity of each FA was calculated as a percentage of the total FA content, then converted to a percentage of the total fatty acid quantity (quantitative relative percentage), with over 97% of the GC peaks identified.

To establish a comprehensive baseline profile of each participant’s functional abilities, sensory processing characteristics, and psychiatric symptoms, a battery of standardized instruments was administered. Core features of autism and overall functioning were assessed using the Childhood Autism Rating Scale, Second Edition – Standard Version (CARS-2-ST) (52). This structured observational tool, appropriate for assessing individuals aged 2 years and older, evaluates 15 domains including verbal and nonverbal communication, emotional response, body use, adaptation to change, and sensory sensitivities. Scores yield a severity index of autism spectrum symptoms, ranging from minimal to severe, and support clinical confirmation of ASD diagnosis. The CARS-2-ST was selected for its focus on behavioural characteristics that remain salient across the lifespan, particularly in adults with high support needs (53). Unlike many adult-specific tools that rely on verbal fluency or self-report, the CARS-2-ST is based on direct observation and caregiver input, making it especially appropriate for individuals with limited expressive language or intellectual disability (54). To assess sensory processing patterns in daily life, the Short Sensory Profile (SSP) was completed by caregivers. This questionnaire evaluates behavioral responses to sensory stimuli across multiple domains (e.g., tactile, auditory, proprioceptive), identifying patterns of hypersensitivity or hyposensitivity that may influence behavior, interaction, and daily functioning (55). Both the CARS-2-ST and the SSP were administered at baseline and the end of the intervention. Psychiatric symptoms and behavioral disturbances were systematically evaluated at all three time points using two validated instruments specifically suited for individuals with severe neurodevelopmental disorders:

The dietary regimen for study participants was systematically modified to incorporate anti-inflammatory and neuroprotective nutritional strategies aimed at improving overall metabolic and behavioral health. The primary objectives were to reduce inflammatory dietary components, optimize macronutrient distribution, and enhance the balance of essential fatty acids, particularly omega-3 and omega-6 polyunsaturated fatty acids (PUFAs).

Fat content in daily meals was calculated by analyzing each recipe’s ingredients based on official nutritional databases (CREA and BDA). Meals were designed to contain total fat levels below 23.3 ± 3.8%, with saturated fatty acids (SFA) limited to 16.6 ± 2.6%. The omega-6 to omega-3 ratio was maintained around 3.2 ± 2.4, and the saturated fat to monounsaturated fat (SFA/MUFA) ratio was controlled at approximately 0.28 ± 0.9. Biscuits provided as snacks contained less than 5% total fat.

Key dietary modifications included the following:

The dietary changes were actively implemented by kitchen staff, educators, and the participants’ families. One author (FC) supervised adherence to the nutritional guidelines throughout the intervention. Under the guidance of a Michelin-starred chef, special attention was paid to food texture, shape, and color to enhance sensory appeal and acceptability for the participants.

Additionally, water intake was actively monitored throughout the day by the center’s caregiving personnel to ensure participants maintained adequate hydration levels consistent with clinical guidelines and individual needs. However, the exact quantities of water consumed were not systematically recorded. This approach was adopted considering the practical constraints of the care environment and the primary focus on ensuring sufficient hydration rather than precise volumetric measurement. Although quantitative tracking of water intake can provide valuable data, in this context, continuous qualitative monitoring by experienced staff was deemed sufficient to prevent dehydration and support overall participant well-being without imposing additional burdens on participants or caregivers.

The study was structured over three distinct time points to carefully monitor both behavioral and neurobiological changes while minimizing confounds related to external supplementation.

This stepwise approach was designed to ensure that the initial improvements could be ascribed solely to dietary changes. It also provided an opportunity to tailor subsequent nutritional support in an evidence-based and personalized manner, thereby enhancing both the scientific rigor and the clinical relevance of the intervention. Overall, this sequential design allowed for a clear differentiation between the effects of natural dietary intake, the isolated impact of dietary modification, and the combined influence of diet plus targeted supplementation on neurobiological and behavioral outcomes in adults with severe ASD.

The distribution of all variables was assessed for normality using both the Shapiro-Wilk and Kolmogorov-Smirnov tests, supplemented by visual inspection of Q-Q plots. Several key variables deviated significantly from normality (p < 0.05), justifying the use of non-parametric methods where appropriate. Spearman’s rank-order correlation was employed to compute the correlation matrix, given its robustness to non-normal distributions and outliers, and its suitability for small sample sizes. Statistical significance of correlations was evaluated at a 95% confidence level. The resulting correlation matrix was visualized using the corrplot package in R; non-significant correlations were left blank to enhance interpretability. For behavioral and psychological measures administered at all three time points (T0, T1, T2), repeated-measures ANOVAs were performed. When the assumption of sphericity was violated (as assessed by Mauchly’s test), the Greenhouse-Geisser correction was applied. Significant main effects were followed up with Dunnett’s post-hoc tests to compare each time point to baseline (T0). The same analytic approach was applied to biological indices and lipidomic data collected at the same time points. For questionnaires administered only at baseline (T0) and post-intervention (T2), paired-sample t-tests were used to evaluate pre–post differences. A significance threshold was set at p < 0.05. However, given the small sample size, statistical power was limited, increasing the risk of both Type I and Type II errors. This limitation inherently constrains the reliability and generalizability of inferential statistics. Accordingly, the primary aim of the study was exploratory: to assess the feasibility of the nutritional intervention, identify potential trends, and generate preliminary data for future, larger-scale trials. In this context, we prioritized descriptive statistics and applied inferential tests only where appropriate and interpretable, while clearly acknowledging their limitations. Substantial inter-individual variability and the absence of a control group further complicate formal hypothesis testing. To avoid overstating findings, we present observed changes over time transparently and cautiously, consistent with the exploratory nature of the study and in line with methodological rigor. All analyses were conducted using RStudio (version 2025.05.1) and all the Figures were generated using GraphPad Prism 8.

The final study sample consisted of seven individuals (6 males, 1 female) residing in one of the living units at the ASP Charitas residential facility in Modena, Italy. Participants ranged in age from 19 to 48 years, with a mean age of 28.8 years (SD = 9.6). All individuals had long-term placement histories at the facility, reflecting the severity of their clinical profiles and the inability to maintain care within the family context. The length of residence varied from 4 to 24 years, with a mean duration of 9.7 years (SD = 6.5). All participants had been diagnosed with ASD requiring very substantial support (Level 3), as defined by the DSM-5-TR, and exhibited significant co-occurring psychiatric conditions or behavioral challenges at the time of the study.

At the onset of the study (Time 0), participants exhibited a high degree of variability in biological and behavioral measures, reflecting a complex interplay between metabolic, inflammatory, and neurobehavioral factors ( Table 1 ).

At baseline, all participants scored above the clinical threshold on the CARS-2-ST, confirming moderate to severe ASD symptoms. The cohort showed high inter-individual variability in symptom severity and behavioral profiles. According to DASH-II and ABC assessments, psychiatric comorbidities such as impulse control problems, manic symptoms, and stereotypic behaviors were common. Irritability and hyperactivity emerged as the most pronounced behavioral challenges, with notable variability across participants. Sensory processing abnormalities were identified in all individuals via the SSP, particularly in the tactile and sensory-seeking domains. Biological assessments revealed no signs of gluten sensitivity, but elevated fecal calprotectin in some cases suggested intestinal inflammation. Most participants exhibited suboptimal metabolic and inflammatory profiles, including elevated HbA1c levels. Notably, IL-6 levels were increased in most individuals, indicating a state of systemic inflammation. Severe vitamin D deficiency was detected in a subset of participants. Lipidomic analyses indicated consistent alterations in erythrocyte membrane composition, particularly elevated stearic, arachidonic acid and an imbalanced omega-6/omega-3 ratio, suggestive of a pro-inflammatory state. All individual data and reference thresholds are presented in Table 1 .

The results of the ABC comparison provide meaningful insights into the changes in autism-related behaviors and symptom severity over time ( Figure 1A ). Post-hoc comparisons were conducted on the scale’s total score using Dunnett’s test to control for multiple comparisons within a single family (two comparisons, family-wise alpha = 0.05). Results revealed a significant difference between T0 and T2 (mean difference = 13.00, 95% CI [1.23, 24.77], p = .0343). Improvements were observed across all subscales ( Figure 1B ), as scores consistently decreased from the beginning to the end of the intervention period. The overall downward trend suggests that dietary intervention may have contributed to a reduction in these challenging behaviors over time.

The analysis of CARS-2-ST compares T2 values with baseline values ( Figure 2A ). A paired-samples t-test revealed a statistically significant decrease in scores between the two time points, t(6) = 4.28, p = .0052 (two-tailed). The mean difference was –3.36 (SD = 2.08, SEM = 0.78), with a 95% confidence interval ranging from –5.28 to –1.44. The effect size was large, with a partial eta squared (R²) of 0.75. This result indicates a general improvement in autism-related behaviors. At T2, all CARS-2-ST subscale scores showed a downward trend compared to baseline (T0), indicating overall improvements across multiple behavioral domains. Importantly, despite these improvements, all participants remained above the CARS-2-ST cut-off score, continuing to exhibit at least moderate levels of autism-related symptoms at T2. This indicates that while the intervention may have reduced symptom severity, it did not lead to a shift below diagnostic thresholds.

Comparison of SSP scores between baseline and T2 ( Figure 3 ) showed a general trend of reduction or stability. The paired-samples t-test comparing total SSP scores between baseline and T2 ( Figure 3A ) did not yield a statistically significant difference, t (6) = 0.85, p = .4293 (two-tailed). The mean difference was –3.43 (SD = 10.71, SEM = 4.05), with a 95% confidence interval ranging from –13.33 to 6.47. The partial eta squared was 0.11, and the pairing correlation was r = 0.66 (p = .0529, one-tailed), not reaching statistical significance. Despite the lack of statistical significance, all participants at T2 had total SSP scores below the clinical cut-off of 154, indicating scores within the normative range for sensory processing. At the subscale level, score trends were either decreasing or remained stable between baseline and T2 ( Figure 3B ).

The analysis of DASH-II items over two time intervals—Baseline to T1 and T1 to T2—reveals notable patterns in both frequency and severity ( Figure 4A ). A repeated-measures ANOVA revealed a statistically significant effect of time on frequency scores, F(1.475, 8.847) = 11.10, p = .0057, with Greenhouse-Geisser correction (ϵ = 0.7373), and an associated R² of 0.65. The pairing was also statistically significant, F(6, 12) = 5.05, p = .0084, R² = 0.47, indicating consistent within-subject differences. Post-hoc Dunnett’s tests showed a significant reduction in behavioral frequency from T0 to T1 (mean difference = 17.86, 95% CI [9.61, 26.11], p = .0015), and from T0 to T2 (mean difference = 15.86, 95% CI [3.95, 27.77], p = .0157). Post-hoc analysis on the severity scale ( Figure 4B ) indicated a significant reduction in severity scores between T0 and T1 (mean difference = 10.43, 95% CI [2.05, 18.81], p = .0210), while the difference between T0 and T2 was not statistically significant (mean difference = 4.43, 95% CI [–7.96, 16.81], p = .5247). Comparing T0 and T2 ( Figure 4B ), most symptoms show an overall reduction in both frequency and severity, despite some fluctuations between T1 and T2. Self-Injury decreases slightly from 4.29 at baseline to 4.14 at T2, after a sharper drop at T1 (2.71), indicating a lasting but partial improvement. Impulse and Control behaviors decline from 22.29 to 16.86, reflecting a sustained reduction over time. Cognitive Deterioration remains stable (9.00 at both T0 and T2), suggesting no significant change in this domain. Anxiety decreases from 3.86 to 3.29, showing modest improvement. Depression initially drops from 13.71 to 7.29 at T1 but rises back to 9.00 at T2, remaining lower than baseline. Mania shows a slight reduction from 9.43 to 8.00. ASD symptoms decrease from 11.43 to 8.43, supporting gradual improvement. Schizophrenia-related symptoms reduced from 1.86 to 1.29. Stereotypies, despite increasing from 7.29 to 8.57 between T1 and T2, remain below baseline levels (8.86). Elimination disorders increased from 1.29 to 2.00, representing a worsening. Eating disorders improved from 3.29 to 0.86. Sleep disorders decline from 1.43 to 0.43, and Sexual disorders nearly disappear, going from 1.71 to 0.29.

This analysis examines changes in key biomarkers across the three time periods ( Figure 5 ). The findings highlight dynamic shifts in biomarker levels, indicating potential metabolic and inflammatory responses to dietary and psychiatric interventions. IL-6 and Fecal calprotectin exhibited non-significant reductions from Baseline to T2 but went back into the accepted range suggesting improvements in inflammatory balance and neurochemical stability. Oxidative Stress Indicator (PCR) showed an increase from T0 to T1, possibly reflecting an adaptive response to initial dietary changes. However, a stabilization trend emerged from T1 to T2, indicating a potential normalization over time. Metabolic Dysregulation (Vitamin D) showed a significant increase from T0 to T1 and T0 to T2. The ANOVA yielded F = 27.84, p = 0.0004. Significant differences were also observed in Insulin Sensitivity (HbA1c) across time points. The repeated measures ANOVA showed a robust effect with F = 38.10, p < 0.0001. Post-hoc comparisons using Dunnett’s test revealed significant increases in HbA1c from baseline (T0) to T1, with a mean difference of 3.71 (95% CI: 2.51 to 4.92; p = 0.0002), and a smaller but still significant increase from T0 to T2, with a mean difference of 1.71 (95% CI: 0.51 to 2.92; p = 0.0116). Minor fluctuations were observed in Lipid Profile (CK), with moderate reductions in the first phase followed by stabilization.

This analysis examines percentage changes in key lipidomic indices associated with fatty acid metabolism and lipid balance across the three time points ( Figure 6 ). The results highlight significant shifts in lipid profiles, which may reflect metabolic adaptations to dietary or psychiatric interventions. DHA and EPA levels showed a moderate increase from Baseline (T0) to T1, followed by a more pronounced increase from T1 to T2. For DHA, repeated measures ANOVA indicated a significant effect over time (F = 13.06, p = 0.0037, Geisser-Greenhouse epsilon = 0.7265, R² = 0.6852). Dunnett’s multiple comparisons test showed a significant increase between T0 and T2 with a mean difference of -1.53 (95% CI: -2.62 to -0.44, p = 0.0124). For EPA, the ANOVA also revealed a significant effect (F = 15.46, p = 0.0019, Geisser-Greenhouse epsilon = 0.7503, R² = 0.7204). Dunnett’s test indicated significant increases from T0 to T1 (mean difference -0.30, 95% CI: -0.49 to -0.11, p = 0.0067) and from T0 to T2 (mean difference -0.43, 95% CI: -0.62 to -0.23, p = 0.0013).

Similarly, the omega-6 to omega-3 ratio showed a significant reduction over time, consistent with the observed increase in EPA and DHA concentrations. Repeated measures ANOVA indicated a significant time effect (F = 10.93, p = 0.0021, Geisser-Greenhouse epsilon = 0.9837, R² = 0.6456). Dunnett’s multiple comparisons test revealed a statistically significant decrease from T0 to T2 (mean difference = 1.714, 95% CI: 0.6350 to 2.794, p = 0.0070). Across the three time points, levels of palmitic, stearic, and oleic acids remained within optimal ranges, showing only minor fluctuations without statistically significant changes.

The correlation matrix presented in Figure 7 shows relationships between neurobiological indices and behavioral measures across the three time points. IL-6 and PCR displayed moderate to strong positive correlations with DASH-II frequency and severity scores. Fecal calprotectin levels correlated positively with ABC scores. HbA1c and fasting insulin levels correlated positively with CARS-2-ST and DASH-II scores. The omega-6 to omega-3 ratio, specifically linoleic acid and arachidonic acid levels, correlated positively with ASD symptom severity and SSP scores. Oleic acid and palmitic acid levels correlated negatively with symptom severity. Vitamin D levels correlated negatively with DASH-II severity scores.