Deficits of facial emotion recognition and visual information processing in adult patients with classical galactosemia

Twenty-two patients, i.e. 58% of the known Swiss CG population ≥ 16 years of age, were enrolled in the study. Thirteen (59%) were females. Mean age was 30 years (SD 11), median 29, range 16–59 years. Of the 22 CG patients all except one were diagnosed through newborn screening. Most of them nevertheless exhibited symptoms of intoxication before treatment initiation. All patients had two known classical mutations (Q188R, K285 N, L195P, H319Q, A320T and M142K), except two patients with one classical and one known slightly milder (R258C) mutations and one patient with the genotype Q188R/L264 V and residual GALT activity of 1.5%. Mean IQ score of the CG patients was 77 (SD 17), the median also 77, and the range 49 to 112. Comparison of patient and control groups did not show significant differences concerning age (p = 0.475) and gender (p = 0.903). The education levels of controls and the parents of the patients were comparable (p = 0.383). But, as expected, they differed significantly between patients and controls (p = 0.028). The three subjects with mild “Duarte” galactosemia were compound-heterozygous for a classical and the Duarte-2 (N314D) mutations. They were clinically normal.

Mean Z-scores of controls were compared to the CANTAB norms by one sample t-test. Overall, the control group was comparable to the normative data cohort of CANTAB (see Additional file 1: Table S1). We therefore compared the test results of CG patients with the results of our control group.

Table 2 presents the descriptive data for all measures of the CANTAB. In the Motor Screening Task (MST), CG patients did not show a significantly longer mean latency or made more errors than controls. In the Paired Associates Learning (PAL) task, CG patients performed worse compared to controls. They needed more trials and made more errors until successful completion of a stage. Patients also made more errors in the stage with six shapes and needed more trials in total. However, none of these measures reached significance in nonparametric testing. In the Spatial Span (SSP) task, CG patients had a significantly shorter span length, which means they could remember shorter sequences than controls. Patients also made significantly more usage errors. These errors are made when the subject selected a box that did not change colour. In the Reaction Time (RTI) task, CG patients performed equally well as controls in the simple-choice part. In the five-choice part movement time was slower for CG patients compared to controls, again without reaching significance, because of the important variation and large overlap with controls. In the Rapid Visual Information Processing (RVP) task, CG patients performed significantly worse in all but one outcome measure. They were less likely to identify the target sequence but did not have more false alarms than controls (Fig. 1). The results of the Emotion Recognition Task (ERT) are displayed in Fig. 2: Recognition of the emotions happiness and sadness was not significantly different between CG patients and controls. However, CG patients had significantly lower percentages for the recognition of the emotions anger, disgust, fear and surprise. They also needed more time to answer than controls.

A random forest model consisting of 5000 trees was trained and validated on the CANTAB data to quantify the relative importance of each outcome measure for the discrimination of group membership (control vs. galactosemia patients; Additional file 2: Figure S1). The higher the MDA (mean decrease in accuracy) of a given measure the more important its contribution to group discrimination. Several global and individual outcome measures of the ERT (including total number correct, mean latency, as well as the emotions surprise and fear) and RVP tasks were most important for group discrimination. Bootstrap estimate of error rate: 35.14%.

A priori significant correlations of ERT and RVP outcome measures are shown in Table 3. When corrected for multiple comparisons (p-value FDR), only three measures, all from the ERT (total correct, recognition of sadness and disgust), correlated significantly with the overall IQ score of the patients but none with their level of education. In contrast, no significant a priori correlation at all was observed with the education of controls (not shown) or the maximum education of the patients’ parents. Note, that no correlation with IQ, educational level or age was found for the ERT fear, which was also important for group discrimination (see above). Two outcome measures, PAL total errors at the stage of six shapes and RTI five-choice reaction time showed a significant correlation with the age of the controls, but not with the age of patients (not shown). No significant correlations between CANTAB outcome measures and any biochemical marker, such as galactose-1-phosphate or residual GALT activity (not shown) were found.

The current study was approved by the local Ethics Committee and all subjects gave informed consent. The International Galactosemia Registry of the European Galactosemia Network (EGN Registry) was implemented in Switzerland in 2015, aiming at the inclusion of all CG patients, most of whom have been diagnosed by new-born screening since the mid-1960ies. All 38 known patients with CG in Switzerland who are ≥16 years of age, were contacted and invited to participate in this study. CG was defined by a known genotype of classical galactosemia or a residual GALT enzyme activity below 10%. Some patients declined to participate in the study because they did not feel well enough (n = 5). One of these suffered from a second condition (Down’s syndrome) and three were born before newborn screening. Others declined because they were not available, mainly for professional reasons (n = 5). For the remaining, the reason is not known (n = 6). The mean age of the non-included patients was 35.3 years (SD 13.1; range 18–59). The final sample consisted of 22 CG patients (59% females). Fifteen controls also completed the CANTAB test battery. They were recruited from a laboratory, administrative and medical staff. It was taken care that their level of education was similar to the level of education of the patient’s parents and male-female proportion was close or identical to the patient cohort, in order to reduce selection bias. In addition, three subjects with mild “Duarte” galactosemia were enrolled. They underwent identical testing as the patient group but were analysed separately. Due to their neuropsychological deficits, patients with CG often achieve lower education than their parents and non-affected siblings. In order to get a measure of the cognitive and psychosocial functioning of the patient families and the controls, of which no full IQ scores were available, we assessed and classified their level of education, as well as of the patients, as follows: “School without qualification”, i.e. either regular schooling not completed or special needs schooling. “School with qualification”, i.e. completed regular schooling and some additional professional training. “Vocational”, i.e. full professional education after obligatory school at age 16 and parallel to high school and college (this is the main educational path in Switzerland, with a good professional standing). “Undergraduate” and “Postgraduate” are the two University levels of the European Bologna system.

The study was part of a larger study conducted at the University Hospital of Bern in Switzerland, which included a full IQ assessment using the Wechsler Adult Intelligence Scale, Fourth Edition (WAIS-IV). Six tasks from the Cambridge Automated Neuropsychological Test Battery (CANTAB) were administered to all subjects including the controls in a session of approximately 60 min using German and French translations of the standardised test instructions and a Windows Surface touch-screen tablet. Thus, each task was explained to the subjects in a thorough and standardized way, and it was made sure that he/she had understood the instructions.

The following tasks were selected (see Additional file 1: Table S2 for a description of the tests):

For reference, see http://​www.​cambridgecogniti​on.​com/​cantab/​cognitive-tests/​ (Cambridge Cognition Ltd., 2017).

Baseline characteristics were presented in a descriptive format, showing mean with standard deviation and median with range for age. The Student’s t-test was applied for mean, the Wilcoxon rank-sum test for median comparison between patients and controls. Frequencies with percentages were shown for categorical variables and the Chi-Squared test was used for comparison. The one sample t-tests was used to compare z-scores of controls to CANTAB normative data. For the comparison of the outcomes of the CANTAB sub-tests we again used the t-test for mean and the Wilcoxon rank-sum test for median, as well as linear models to compute p-values corrected for the influence of age, gender and the maximal educations of parents. As multiple statistical tests were performed, these p-values were adjusted using the method of Benjamini and Yekutieli to reduce the false discovery rate. The relative importance of outcome measures for predicting CG status was assessed by a random forest model consisting of 5000 trees, implemented in the randomForest R package. The random forest bootstrap estimate of error rate was 35.14%. The Spearman correlation was used and p-values were computed using Spearman’s rho statistics. All analyses were performed in version 3.4.1 of the R statistical environment.