Data were collected from the Objective Diagnostic Markers and Personalized Intervention in MDD Patients (ODMPIM) study, which was carried out between December 2013 and December 2016. The participants were part of a clinical trial and a detailed description of the ODMPIM study protocol can be found elsewhere (22). Briefly, the participants were recruited from nine clinical sites at nine top tertiary hospitals (seven within academic settings and two in clinical practices) located in six cities. The number of cases recruited by each clinical site is shown in Supplementary Table 1. The details of inclusion and exclusion criteria for each group are listed in Supplementary Table 2. Finally, this study included 1,629 Chinese participants (1,130 patients with a current diagnosis of MDD and 499 healthy controls) aged 18–55 years. The validated Chinese version of the Mini-International Neuropsychiatric Interview (MINI, Version 5.0) was used to diagnose current MDD. The 17-item Hamilton Depression Rating Scale (HAMD-17) was applied to assess the severity of depression over the past 2 weeks. The study was approved by the Ethics Committee of Peking University Sixth Hospital (Approval No. 2013-29-1). Before respondents were interviewed, written informed consent was obtained from them.

All assessments were acquired at baseline prior to randomized treatment in the ODMPIM study. The measurement scales included the Childhood Trauma Questionnaire (CTQ), the Simplified Coping Style Questionnaire (SCSQ), and a series of neurocognitive tests.

After systematic data cleaning, the categorical variables were presented as frequencies and percentages. For the continuous variables, a Kolmogorov–Smirnov (K-S) test of normality was performed, and the data were expressed as the mean (standard deviation) if they were normally distributed and the median (interquartile range) if they were not. An independent-sample t-test, Mann–Whitney U test, Chi-square test, and Fisher’s exact test were used to compare the demographic data between MDD and HC groups, as appropriate. The differences in CM, personality traits, and coping styles between the two groups were tested by logistic regression and linear regression controlled by age, sex, and education levels. Significance levels was considered pre-set to 0.05 based on two-tailed tests.

In total, 1,130 patients with MDD and 499 healthy controls participated in this study. Twenty-one patients with incomplete demographic data (age, work status, marriage, and education) were excluded. Finally, 1,608 participants were included in this study. 53.8% of MDD patients were assessed as having moderate depression by HAMD-17, and the median score was 21.0 (17.5, 24.0). The median and interquartile range of illness duration were 6.0 (3.0, 13.0) months. The daily antidepressant doses were converted to fluoxetine equivalent doses using the following equations: fluoxetine 40 mg/d = citalopram 40 mg/d = escitalopram 18 mg/d = sertraline 98.5 mg/d = paroxetine 34 mg/d = fluvoxamine 143.3 mg/d. The median and interquartile range of the daily dosages after converting to fluoxetine equivalent doses was 23.53 (20.28, 33.33) mg/d. All the healthy controls had lower HAMD-17 scores, with none of them being assessed as a depressive state. The mean age of the MDD and healthy groups was 39.40 and 33.95, respectively. There were significant differences in social-demographic information between the two groups. Details are provided in Table 1.

After controlling by age, sex, and education levels, CM’s prevalence rates and severity scores were higher in MDD patients than in healthy controls, except for emotional abuse. Besides, scores on all personality dimensions differed significantly between the MDD and healthy groups, except for the lie scale. The MDD group had lower positive coping style scores and higher negative coping style scores. See Table 2 for further details.

For the relationships among different CM types, emotional abuse, emotional neglect, physical abuse, and physical neglect formed quadrangle connections in both the HC group and the MDD group. Personality dimensions showed associations with physical neglect, emotional abuse, and emotional neglect. All types of CM (excluding SA) showed an association with coping style (Figure 1). The most substantial edges were between SA and L (0.343), EN and P (–0.206), PN and negative (0.158), and EN and N (–0.149) among the healthy control group and between EN and positive (–0.103), PN and P (–0.092), EN and N (–0.081), and EA and E (0.069) among the MDD group. Bootstrapped confidence intervals for the edges were shown in Supplementary Figures 1, 2.

Of all adversities, EN showed the highest centrality measures (Figures 2, 3), indicating it lay on many shortest paths between every combination of two other nodes in the network. This suggested that EN played a vital role in different routes through the network. PN had a high level of closeness, meaning it had a short average distance from all other nodes, suggesting an influential role in the network. Emotional and physical neglect showed the highest levels of strength, indicating that they had many or strong connections with other nodes in the network. Among the healthy control group, the centrality measures were much lower for neglect. PN, EN, and EA showed higher closeness and strength. Comparisons of the two networks based on NCT results showed that the structure of the networks was variant between groups (M = 0.28, P = 0.04), and the global network strength test similarly revealed no significant differences in the weighted sum associations of trait facets (S = 0.44, P = 0.41).

Latent class analysis revealed that three classes provided the best fit statistics. From the results of multiple-group latent class analysis, it could be seen that the latent class probabilities were different between the MDD and HC groups. Information regarding model fit statistics and selection criteria were shown in Supplementary Table 3. For the MDD group, the LMR-LRT and BLRT were significant for the two- and three-class models and marginally significant for the four-class models. The three-class solution had the lowest information criteria (AIC, BIC, and SS-BIC) and demonstrated high entropy (0.917), suggesting good class identification. Table 3 depicted the observed class membership and endorsement frequencies. Class 1 was termed “neglect,” characterized by very high endorsement rates for PN and EN. Class 2 was termed “normal.” Class 3 was termed “abuse,” characterized by prominent symptoms of EA. The proportions of these three classes were 15.63, 80.89, and 3.47%. For the HC group, the two-class solution had the lowest information criteria and demonstrated high entropy (0.920). The LMR-LRT and BLRT were all significant in two-, three-, and four-class models. The proportions of these two classes were 8.07 and 91.93%.

We compared the differences in the T-scores for each cognitive domain and each test among different classes and between two groups, as shown in Table 4. Unexpectedly, the MDD group had worse cognitive function on all dimensions than the healthy group. In general, participants who had a history of maltreatment tended to perform more poorly on the Continuous Performance Test, the Hopkins Verbal Learning Test, the Brief Visuospatial Memory Test, the Digit Symbol Coding Test, and the Stroop Color and Word Test. The results revealed significant differences between the different classes in cognitive functioning in five domains. SNK test results showed that compared to the other two classes, the neglect class had lower cognitive function scores in learning, processing speed, and executive function. The abuse class had the lowest scores in attention and memory (Figure 4).

The network analysis showed that CM, especially childhood neglect, played an essential role in personality traits and coping styles. Neglect had the highest number and strength of the connections, which meant it might have intermediary effects or be more closely related to all other aspects. The moderating effect of neglect had been found on life satisfaction, depressive symptoms, cognitive functions, treatment outcomes, and suicide risk (40). Social learning theory emphasizes the importance of observing and modeling the behaviors, attitudes, and emotional reactions of others. Children spontaneously imitate their parents to learn reactivity, and disciplinary methods are affected by the family atmosphere (41).

In our study, participants with neglect experience tended to exhibit hyper-reactivity to negative emotions as a result of acquiring negative affect and poor regulation strategies from their parents (42). However, neglect was often ignored in previous studies (43). Physical and sexual abuse is easier to identify and tends to attract more attention from social workers and researchers, even the media (44, 45). Neglect is much more challenging to identify; it is characterized by a lack of care and protection rather than direct explicit abuse. In addition, because children with neglect might be subjected to other forms of abuse simultaneously, the independent impact of neglect was often ignored by researchers. This is particularly concerning because studies suggest that the consequences of childhood neglect might be as severe as those of other types of maltreatment. In the long term, neglect is at least as damaging as physical or sexual abuse, but it receives the least public attention. Surprisingly, sexual abuse did not show significant associations with any of the personality dimensions. Previous findings also showed inconsistent results. This contradiction might be caused by the biases in self-reports of sexual abuse. When answering questions related to sex, there might be problems such as forgetting, denial, misunderstanding, and embarrassment that are likely to lead to the under-reporting of the sexual abuse of children (46). Besides, children’s sex education in China is not universal, which could lead to poor ability to recognize sex abuse.

The differences in network structure between groups might indicate the pathological mechanisms of MDD. EN was directly related to P in the healthy group, and PN was directly related to N. However, this relationship disappeared in the MDD group. The person who could buffer the impact of stressful events such as CM on adult personality traits might not develop MDD in adulthood. Although the etiology and pathogenesis of depression are still unclear, CM is regarded as a critical risk factor for depression. This might partly explain why not all children with CM become depressed. Individuals’ personality traits may significantly influence the association between CM and later-life depression (20).

Children who were exposed to one type of maltreatment were often exposed to other types. The results of LCA also showed these strong connections, which were in line with previous literature. Furthermore, the particularly strong connections between physical and emotional CM (both abuse and neglect) confirmed that physical CM was often accompanied by emotional CM. Though the MDD group was divided into three classes and the healthy group was grouped into two classes, the kinds of classifications were still similar. PN was in a dominant position in class 1, accompanied by EN, in both populations. This also demonstrated that the co-occurrence of different types of CM was more common within abuse (i.e., PA and EA) or neglect (i.e., PN and EN) than between abuse and neglect (i.e., PN and EA). The main difference was in Class 3 of the MDD group. EA was in the dominant position, accompanied by PN and PA. Showing that neglect and abuse co-occur more commonly in MDD patients compared to the healthy population. It should be highlighted here that neglect and abuse likely represent the two extreme polarities of CM. Neglect is the most relevant form of maltreatment “by omission,” in which the child is deprived of the basic needs of protection, care, and love from caregivers. On the contrary, abuse represents “by commission”; the caregivers degrade, humiliate, and terrorize their children to show their power or control over them. According to Infurna’s meta-analysis (47), both types may develop a more negative self-model, becoming prone to internalizing symptoms, which may easily foster depression later in life.

This study investigated the relationship between latent classes of CM and adult cognitive functioning. A systematic review was previously conducted evaluating the evidence for an association between maltreatment and cognition in children under 12 years. However, there has been little research on adult cognitive function. Significant differences were found in five domains among the different classes (i.e., attention, learning, memory, processing speed, and executive function). CM (both abuse and neglect) was associated with the development of cognitive functioning. There are two possible explanations for this. One is that children with maltreatment could have potential differences in brain anatomy or patterns of disrupted cognitive functioning compared to those without (48). The other might be the equivocal causation between CM and cognitive function (49).

Furthermore, the two core dimensions of CM—abuse and neglect—might influence different dimensions of cognitive function. Our results showed that the neglect class performed more poorly in learning, processing speed, and executive function, while the abuse class performed more poorly in attention and memory. This might be because different CM experiences were related to the development of two underlying neural processes of cognition. Specifically, neglect was a more prominent predictor for developmental changes in insula-dACC activation during risk processing, while abuse was a more prominent predictor for developmental changes in frontoparietal activation during cognitive control (50). However, the association between maltreatment and adulthood cognition is a complex and pendent problem. Notably, when considering causality, the evidence from cross-sectional studies is weak, as the direction of causality could be from maltreatment to cognitive problems or vice versa (51). Danese et al. (49) found that even though there was impairment in cognitive functioning among those exposed to CM, this impairment was explained mainly by cognitive difficulties that pre-dated CM exposure and confounding genetic and environmental factors. As a cross-sectional study, the results here could only describe the difference. Longitudinal studies are needed in the future to explore the mechanics behind it.