This is the first cortical thickness study for non-NPSLE patients that uses both the SBM and ROIs analysis. We also performed correlation analyses on the cortical thickness abnormalities of different brain regions obtained by the ROIs method and a large number of clinical indicators. Finally, we constructed and analyzed the SCNs of cortical thickness in non-NPSLE patients for the first time, and discovered structural network abnormalities in non-NPSLE patients. Specifically, we determined that non-NPSLE patients exhibit: (1) abnormal cortical thickness: the results of the two analytical methods both show that non-NPSLE patients have a wide range of brain regions with cortical thinning, when compared with HCs (2) multiple clinical indicators, disease activity and mental scale results are in correlation with these abnormal brain regions (3) changes in global network measures include the improvement of clustering coefficient, the breakdown of small-world attributes, fluctuating regional network measures in some of brain regions and changes in both the number and distribution of different network hubs. The above findings may provide meaningful information to deepen our understanding of SLE brain damage and explain the underlying mechanism of non-NPSLE associated early neuropsychiatric abnormalities.
Cortical thickness is one of the most important indicators of brain structure analysis. In this study, before the appearance of obvious neuropsychiatric symptoms and conventional MRI abnormalities in non-NPSLE patients, extensive cortical thinning of the bilateral cerebral hemispheres was observed, it suggests that the cerebral cortex of non-NPSLE patients have undergone significant subclinical changes before evolving into NPSLE. Zivadinov et al.[43] conducted multimodal neuroimaging studies on lupus patients and found that cortical atrophy was the most relevant measurement index for central nervous system involvement in SLE. Similar SBM studies for SLE also found that SLE patients have thinner cortical thickness in multiple brain regions[9, 19, 20, 44]. Many previous studies[10, 45–47] have also verified the cerebral cortical atrophy in SLE patients by VBM and other different technical modalities, therefore cortical thickness can be regarded as one of the imaging biomarkers for structural changes in SLE patients.
Although cortical atrophy in SLE patients has been extensively reported, the mechanisms of brain injury related to SLE are complex and the exact mechanisms have not been elucidated[48, 49]. Present understanding in the matter is that multiple interrelated mechanisms attributed to underly SLE related brain damage include blood-brain barrier dysfunction, vascular inflammation, thrombosis, vascular occlusion caused by atherosclerotic changes, neuroendocrine imbalance, tissue and neural damage mediated by autoantibodies (such as anti-ribosome P0 antibody, ACL) and proinflammatory factors (such as IL-1, IL-6, IL, 8, TNF - α) combine to produce neuronal loss[50–53]. And progressive diffuse neuronal loss will eventually lead to the atrophy of, and consequently, thinning of cerebral cortex observed by structural magnetic resonance in this study.
In this study, the SMB and ROIs methods were applied on the same group of subjects to compare cortical thickness based on the DS atlas. SBM analysis with post-correction (Monte-Carlo simulation cluster analysis with 1,000 permutations, cluster-wise threshold of P < 0.05 and cluster-forming threshold of P < 0.001), revealed persistent significant cortical thinning in 34 clusters in non-NPSLE, which reflects the severity and diffuse nature of non-NPSLE related brain damage. Prior SBM-based cerebral cortex studies in lupus patients have used FreeSurfer default DK atlas without exception. There has been no previous study separately on the non-NPSLE subgroup of SLE. Some brain regions with thin cortex identified here are well reported, such as the left supramarginal gyrus of NPSLE is thinner than control groups[20], the left superior temporal gyrus of SLE patients with episodic memory deficit is thinner than control groups[19], and the left superior parietal cortex of NPSLE is thinner than non-NPSLE[9], but newer abnormal brain areas have been discovered for the first time. We consider that this perceived discrepancy can be explained by differences in study population and the use of brain anatomy atlas. Current and prior studies suggest that the patterns of cortical involvement may be unique to subgroups of NPSLE, however specific differences remain to be verified. In ROIs analysis, we found more abnormal brain regions (57 regions in bilateral hemispheres), all of them showed cortical thinning in the non-NPSLE group, and 19 brain regions overlapped with SBM results. Incomplete congruency between the two results reflects that the anatomical regions defined according to the structural boundary may not correspond to the pathophysiological indicators of SLE patients' cortex. Similar results have also been found in currently depressed study[54]. SBM is perhaps the more suitable, of the two methods, for detecting cortical thickness alterations that do not fit perfectly into predefined regions, but the superiority of one method over the other in detecting cortical abnormalities in SLE can only be established by further verification.
In this study, course of the SLE was found to be negatively correlated with cortical thinning in some brain regions; and no positive correlation were found, which also reaffirms the results of previous studies related to SLE. Cerebral injury associated with SLE seems to be a continuous process, and analysis of cortical thickness can be used as a clinical indicator for the evaluation of brain injury in SLE with varying disease courses Similar conclusions have been reported in studies of type 2 diabetes[55]. Quantitative analysis of urine protein is the main clinical indicator for detecting lupus nephritis. Although the mechanism is not fully understood, the currently accepted view is that immune-mediated inflammation is the main cause of lupus nephritis[56]. Cortical thickness of some brain regions is negatively correlated with the range of proteinuria, suggesting a certain degree of correlation between lupus nephritis and lupus brain damage. Correlation analysis also found that the cumulative dosage of several commonly used immunosuppressants is positively correlated with the average cortical thickness of multiple areas of the brain, suggesting that these immunosuppressants may have a potential protective effect on cortical thickness. Our previous study also found that compared with the patients had never received immunosuppressive therapy, immunosuppressive therapy of SLE patients with average whole brain white matter volume tends to increase[57], and that the immune inhibitors nerve protection mechanism may be the result of the reduced nerve injury due to vasculitis. The pros and cons of long-term immunosuppressive therapy need to be clarified by more prospective studies.
The presence of multiple autoantibodies in serum is one of the most prominent features of SLE, and some autoantibodies may also play a key role in SLE-associated central nervous system injury. Serum levels of brain reactive antibody in SLE patients is unrelated to neuropsychiatric symptoms, but their levels in cerebrospinal fluid are proven to be significantly related to neuropsychiatric symptoms. Therefore, the destruction of BBB is considered to be crucial for autoantibodies to enter the central nervous system. At present, the well-studied brain reaction autoantibodies associated with SLE include anti-N-methyl-d-aspartate receptors (NMDAR), anti-ribosome P0, microtubule associated protein 2 (MAP-2), matrix metalloproteinase 9 (MMP-9), RO anti body [58], anti U1RNP and anti-phospholipid (APL) antibodies[59–61]. This study showed that serum ANA, LAC and anti-nucleosome antibody were negatively correlated with cortical thickness in some brain regions, confirming their brain damaging effect. However, serum anti-SSA52KD antibody and anti-P0 antibody were positively correlated with cortical activity in some brain regions, which was contrary to previous knowledge. At present, there are few studies on the exact relationship between antinuclear antibody spectrum and brain structural abnormalities. The specific mechanism by which autoantibodies exert effects on brain structure changes in SLE needs to be determined by future cerebrospinal fluid antibody research and animal experiments.
SLEDAI-2k has important value in the overall assessment of lupus disease activity, and has reached a consensus among experts in the field of lupus research; it is widely used in clinical evaluation and scientific research of SLE. This study did not find a significant linear correlation between SLEDAI and cerebral cortex thickness, and none of the existing SLE cortical studies involved SLEDAI. In the diffusion-weighted magnetic resonance study, the SLEDAI-2k score also showed no correlation with mean diffusion or fractional anisotropy[62], while SLE brain studies have reported SLEDAI score and regional WM volume for the right internal capsule and left internal capsule[57]. On the one hand, these differences indicate that the disease activity in a certain period may indeed be irrelevant to the changes in brain structure caused by the long-term disease process. On the other hand, this again reflects the ambiguity of the current mainstream SLE disease activity assessment system for neuropsychiatric assessment and its incompatibility with the current rapidly changing advanced neuroimaging technology. The importance of magnetic resonance in the diagnosis and evaluation of neuropsychiatric lupus has also been valued by rheumatologists[63].
Previous studies have observed decreased white and grey matter in SLE patients with cognitive dysfunction in comparison to those with moderate cognitive impairment[45]. Our previous research also found that MMSE score is positively correlated with gray matter volume[10]. In this study, 63% of the subjects had normal cognitive function; whereas, 32% had mild and 5% had moderate cognitive impairment, no subject has severe cognitive impairment. There is a positive correlation between MMSE grade and the cortical thickness of right G&S_cingul-Ant. The cingulate gyrus is part of the limbic system of the brain, and its functions involve emotion, learning and memory. Clinical studies have shown cingulate abnormalities in many cases, including schizophrenia, depression, post-traumatic stress disorder, mild cognitive impairment, and Alzheimer's disease. Cognition is the functional result of learning and memory process produced by the interaction of various neurotransmitters, transcription factors, cytokines and chemokines between neurons, astrocytes, glial cells and immune cells. Cognitive dysfunction is a common phenomenon in SLE[64, 65], however, its pathological mechanism still needs additional study. Some studies have found that there are multiple factors related to anxiety and depression in SLE, including certain specific autoantibodies, nerve damage, the presence of rash, the concentration of certain cytokines, pain and disability caused by pain, socioeconomic status, and factors related to socioeconomic status, such as reserve capacity and psychological resilience[66–68]. These different findings suggest that depression and anxiety in lupus patients may be mediated by a complex mix of biosocial and environmental factors. This study found that some non-NPSLE patient have anxiety and/or depression, as shown in Fig. 1, and HAMA and HAMD grades were positively correlated with cortical thickness in some brain regions, respectively. In a structural magnetic resonance study of primary anxiety and depression, there was no significant correlation between anxiety symptoms and brain structural indicators, while depression symptoms were related to the thinner cortical thickness of some brain regions[69]. A recently published large-scale analysis combining global data shows that the brain structure of patients with generalized anxiety disorder has not changed significantly[58]. At present, in SLE, the relationship between anxiety, depression and cerebral cortex thickness is still unclear, and more targeted research is needed to verify it in the future.
In recent years, brain connectivity has been widely used to study the pathological mechanism of brain damage. For SLE, studies have also found abnormalities in functional connectivity[42] and white matter connectivity[70, 71]. SCNs studies on SLE cortex are yet unavailable. The first part of this study also shows that changes in cortical thickness in a non-NPSLE brain are widespread, and not confined to a few brain regions. Therefore, in order to explore the multivariate network relationship between different neuroanatomical regions in the context of non-NPSLE, we further conducted a structural covariant network analysis based on cortical thickness.
Between groups comparison of global network measures primarily found that except for Dmin, the clustering coefficient of the HCs in other densities was lower than that of the non-NPSLE patients (Figure. 4B), while other measures were not significantly different (Figure. 4). Briefly, the clustering coefficient is the ratio of edges between nodes in a neighborhood divided by the number of edges that may exist between them; it represents the degree of interconnectivity between network nodes and neighboring nodes. The changes in clustering coefficient of SCNs in different diseases are heterogeneous, for example, it is increased in tinnitus[72], decreased in type 2 diabetes mellitus[73], whereas, no significant change is seen in vertically infected HIV adolescents [74]. The extensive reduction of cortical thickness and the enhancement of the interaction between abnormal cortex may be characteristics of non-NPSLE brain damage. Presence of initial brain lesions and their progression needs to be explored in future studies. The small-world attribute reflects the basic elements of human brain’s information processing system: functional separation and functional integration; the former refers to the ability of closely connected brain regions to process information, and the latter refers to the ability of arbitrarily distributed brain regions to transmit information. In this network structure, local adjacent brain regions are closely connected, and a small number of connections are created between any two brain regions for rapid communication. The balance between local information processing and whole brain transmission is achieved, which not only meets the efficiency of functional classification and functional integration, but also reduces the cost of maintaining efficient communication[75, 76]. In this study, the non-NPSLE group does not follow the small-world attribute in a few densities (small-world index༜1, see Figure. 3I), reflecting the sub-optimization of non-NPSLE SCNs.
For comparison of regional network measures between groups, we performed AUC (density range of 0.38–0.5 with an interval of 0.01) analysis on normalized clustering coefficient, degree, betweenness and local efficiency of each brain region in the two groups (see Figure.5). We found that some brain regions displayed increased or decreased values for the above given measures, reflecting the extensive and obvious changes in the SCNs attributes of the non-NPSLE cortex in the local brain regions. DS atlas, utilized to construct the network in this study, is divides the cerebrum more finely and into more regions than DK atlas adopted by many other similar studies done on other diseases. [39, 77, 78] This explains why the network measures found in this study have changed in more nodes. On the other hand, perhaps like the reduction in cortical thickness, the abnormalities of regional SCNs in large areas of the brain and the compensatory changes in corresponding brain areas are the characteristics of non-NPSLE. This will also require SLE magnetic resonance data from other research centers in the future to assist in verification.
The hubs of non-NPSLE and HCs are different both in number and location.
Of the 7 hubs of non-NPSLE, 3 are located in the frontal lobe, 2 are in the anterior cingulate cortex, and 2 are in the temporal lobe. Of the 5 hubs of HCs, 2 are in the supramarginal gyrus, 2 are in the temporal lobe, and 1 are in the insula. A main area for hubs increases, studies have found that the anterior cingulate cortex and temporal lobe are related to emotions and learning[79–81]. However, the pathological mechanism of the changes in these locations has not yet been elucidated in SLE. The increase of hubs in these regions may be a compensatory supplementation. No common hubs were found between non-NPSLE and HCs. The increased number of hubs associated with changes in the location may be a feature of the non-NPSLE cortical thickness covariant network. These findings may also be caused by individual differences between subjects.
This study also has some limitations. First of all, the field strength of the magnetic resonance scanner used in this study (1.5T) is lower than that currently mainstream in brain imaging research (3.0T), so our results may be biased due to the smaller signal-to-noise ratio. We actually started this research 10 years ago and continued the same 1.5T MRI scanner and scanning parameters to build the database. In this study, we conducted strict quality control, including manual visual inspection and software quality control. We added detailed quality control information in the supplementary materials. We also hope to use new magnetic resonance scanners in the future to obtain more accurate results. Secondly, cross-sectional design of this study makes it impossible for us to describe the dynamic changes in non-NPSLE cerebral cortex and SCNs along with disease progression, this can only be achieved with longitudinal research. Thirdly, when performing SCNs analysis, each group has only one network, instead of each subject having a separate network, so we cannot test the relationship between network parameters and clinical measurements.