Synapse loss in the prefrontal cortex is associated with cognitive decline in amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) is the most common form of motor neurone disease (MND), resulting in progressive upper and lower motor neurone dysfunction and loss. This leads to progressively severe muscle wasting and most patients succumb due to respiratory failure, with a mean survival time of 3 years post diagnosis [57]. Due to the striking, and in many cases rapid, deterioration of motor function, ALS was historically considered a homogenous disease of the motor system. However, both clinical and post-mortem studies increasingly describe ALS as a heterogeneous disorder, with around 50% of patients exhibiting neuropsychological deficits due presumably to breakdown of non-motor areas of the brain [49]. ALS is thought to lie on a disease spectrum with frontotemporal dementia (FTD). These seemingly disparate disorders share both genetic and pathological features [30] with considerable clinical overlap. An estimated 15% of FTD patients present with motor decline meeting the criteria for ALS diagnosis and 15% of ALS cases show severe cognitive and or behavioral decline, meeting the FTD diagnostic criteria [24, 30, 49]. However, a further 30–40% of ALS patients show milder, more specific cognitive and/or behavior changes (ALS cognitive impairment—ALSci/ALS behavioral impairment ALSbi [49]) that does not meet FTD diagnostic criteria but can still affect daily life [20]. Changes include executive dysfunction, decreased verbal fluency, language impairment and social cognition deficits, which are thought to be mediated by regions of the frontal cortex and are similar to mild FTD-associated impairments [20]. Behavior change is predominantly one of the apathies, while other features of FTD are also seen, including disinhibition and loss of sympathy/empathy [20]. Functional and structural imaging studies support the presence of frontal deficits [11, 34, 47, 56], and specifically relate dysfunction of the dorsolateral prefrontal cortex [including Brodmann Area 9 (BA9)] to the cognitive profile in ALS [1, 2]. It is vitally important that we understand the underlying neuropathology driving these cognitive deficits, as cognitively impaired patients not only endure the personal burden of both motor and cognitive dysfunction, but also exhibit a shorter survival time from symptom onset [15, 58].

Synapse degeneration is a common feature in a large number of neurodegenerative diseases and often plays a role in pathogenesis, rather than simply occurring as a secondary effect of neuron loss [22]. For example, synapse loss remains the strongest neurobiological correlate of cognitive decline in Alzheimer’s disease (AD [54]). In AD, synaptic accumulation of key pathological forms of tau and beta-amyloid are thought to drive synaptic breakdown [48]. Likewise, studies in ALS mouse models have suggested that synapse defects in motor regions of the cortex are an early/pre-symptomatic feature of the disease [18, 43]. However, to date, few studies have attempted to assess the contribution of synaptic change to pathology of non-motor cortical regions in the brain of human ALS patients and none using direct measurement in patients with defined ALS with cognitive impairment.

By combining the high-resolution imaging techniques of array tomography and transmission electron microscopy (TEM), we have examined synapses in the frontal cortex and motor cortex of cognitively unimpaired and cognitively impaired ALS (ALSci) patients, to define the relationship between synaptic breakdown and cognitive impairment. Cognitive profiling was performed using the Edinburgh Cognitive and Behavioural ALS Screen (ECAS) [3], a short multi-domain assessment designed for people with physical disability which has been validated against extensive neuropsychology [38]. Other post-mortem measurements such as cortical thickness, pathological burden and gliosis were recorded to ensure a comprehensive profiling of each case.

Transmission electron microscopy (TEM) analysis of the prefrontal cortex [Brodmann Area 9 (BA9); ~ 150 synapses per case] from 5 control and 20 ALS cases (Table 1; cohort 1) revealed a significant decrease in synaptic density in the ALS cohort (Fig. 1a–c). In addition, there was a significant increase in the number of electron-dense, degenerating pre-synaptic terminals in ALS cases compared to controls (Fig. 1b, d) [19].

To expand our synaptic analyses, we next examined a largely independent cohort (Table 1; cohort 2a) of subjects using higher-throughput array tomography. Analysis of over 1 million synapses (~ 36,000 synapses per case) from 14 controls and 29 ALS brains using array tomography confirmed our initial TEM findings revealing synapse loss in BA9 of ALS patients (2-tailed unpaired t test; p = 0.013; Fig. 1e–g). Post-mortem interval (PMI) did not influence synapse density (Pearson r = − 0.21, p = 0.17). Importantly, the loss of synaptic density was not accompanied by any detectable gross atrophy of BA9 (2-tailed unpaired t test; p = 0.68; Fig. 1h). Furthermore, stereological estimates of neuron density counts from 5 cases with the highest and lowest synapse densities were not significantly different (Low: n = 5, 23,370 ± 2756 neurons/mm³; High: n = 5, 28068 ± 2094 neurons/mm³. Unpaired 2-tailed t test; p = 0.22), suggesting that synapse loss was occurring in the absence of concomitant, overt neuronal loss.

We then performed some exploratory analyses to assess the influence of numerous genetic and pathological factors that may associate with increased synapse loss. Genetic screening revealed that 6 ALS cases had a hexanucleotide repeat expansion in C9orf72, 3 cases had SOD1 variations and 2 had NEK1 variations (Supplementary Table 1). However, we did not observe any significant association between lower synapse density in BA9 and distinct genetic mutations in ALS with this sample size (Fig. 1g; Supp. Fig. 1).

Pathological changes in disease-associated proteins are strongly implicated in driving synapse degeneration in dementias [48]. Thus, we next assessed the presence of disease-associated pathological aggregates in formalin-fixed paraffin-embedded (FFPE) sections from BA9 and their association with synapse loss in array tissue from the same case. We found no association between the presence of beta-amyloid or pTau pathology in BA9 and synapse loss (Supp. Fig. 2). We also found no association between the presence of pTDP-43 aggregates in BA9 and synapse loss (Supp. Fig. 3). However, it is known that ALS patients with SOD1 mutations do not express pTDP-43 pathology [32]. When we remove our three SOD1 + ve cases from the analysis, we find a significantly lower synapse density in ALS cases with pTDP-43 pathology (Supp. Fig. 3). Due to this potential association and to the literature from Alzheimer’s disease in particular, indicating that pathological protein accumulation at synapses can drive synapse degeneration [48], we used array tomography to assess the presence of pTDP-43 at the synapse. We observed pTDP-43 pathology in all of the 9 ALS cases screened, often exhibiting as large cytoplasmic aggregates or neuritic threads (Fig. 1i). Interestingly, we also noticed small pTDP-43 puncta in the neuropil of our 9 cases, which in a subset of synapses co-localized with synaptic markers (Fig. 1i, j).

Twenty-three patients from whom we collected tissue for array tomography had previously been cognitively profiled using the Edinburgh Cognitive and Behavioural ALS Screen (ECAS). These individuals (Table 1; cohort 2b) were classified as unimpaired (ALS) or ALS with cognitive impairment (ALSci) according to established threshold scores for ECAS Total or ALS-specific ECAS results [3]. The scores revealed 16 unimpaired cases and 7 impaired cases [70% unimpaired, 30% impaired (Fig. 2a)]. Comparing ECAS data with synaptic data, we found a significant difference between our three groups in synaptic density and that the cognitively impaired group had a significantly lower synapse density in BA9 than controls (Fig. 2b). We found no significant difference between the ALS unimpaired cohort and either controls or ALSci, most likely due to two cases with very low densities in the unimpaired group (Fig. 2b). When we plotted ECAS score versus synapse density, we found no significant correlation between these data (Pearson r = 0.31, p = 0.15), again due to the influence of the two unimpaired cases with low synapse density mentioned above (Supp. Fig. 4). One potential confounding factor is the length of time between cognitive testing and post-mortem synapse analysis. However, we found no correlation between synapse density at post-mortem and time since cognitive testing (Pearson r = 0.043, p = 0.85). Although there was no association between genetic status and synapse density or cognitive impairment, it was interesting to note that 3 of our 4 genetically screened cognitively impaired cases were positive for C9orf72 repeat expansions (Fig. 2b), warranting further study of the effects of C9orf72 expansions, as more cases prepared for these techniques become available. Cortical thickness did not vary between the groups (ALS, ALSci, Controls) (Fig. 2c), suggesting that atrophy and overt neuron loss were not driving cognitive decline. We found no association between age and ECAS score (Supp. Fig. 5) or cortical thickness and ECAS score (Supp. Fig. 5). Furthermore, the presence of beta-amyloid or pTDP-43 pathology in BA9 did not associate with lower ECAS score (Fig. 2d, f). Interestingly, although the presence of pTau pathology did not associate with a lower mean ECAS score (Fig. 2e), pTau + ve cases were predominantly ALSci (5 of the 7 pTau + ve cases were ALSci), whereas pTau-ve cases were predominately unimpaired ALS cases (12 of the 14 pTau-ve cases were unimpaired ALS; Fisher’s exact test: p = 0.005). This suggests that pTau pathology may associate with cognitive impairment and warrants further study once more cases become available.

Eleven of the ECAS-screened cases also underwent the ECAS behavior interview, which was completed by an informant/carer in vivo. Six cases were unimpaired and five were classified as ALS with behavioral impairment (ALSbi), with changes including apathy, lack of empathy and behavioral disinhibition. Analysis revealed no difference in BA9 synapse densities between the three groups of controls, ALS and ALSbi (One-way ANOVA; F = 1.93; p = 0.17).

Together, these data suggest that synapse density is a robust indicator of ECAS score and cognitive performance in ALS.

To address whether synapse breakdown, in the absence of cortical atrophy, is a global feature of pathology in the ALS brain, or a specific event occurring in the prefrontal cortex, we employed an array tomography approach to analyze ~ 750,000 synapses from the primary motor cortex. Comparing synapse density in 7 controls versus 25 ALS patients, we discovered no difference in synapse density (Fig. 3a–c). Grouping the BA4 synapse values based on cognitive status did not reveal any difference in synapse density between the groups (Fig. 3d). However, there was significant cortical thinning of the motor cortex in ALS patients compared to controls (Fig. 3e). When the ALS group was divided into unimpaired ALS and ALSci, it was found that both groups exhibited significant cortical atrophy compared to controls (Fig. 3f). Furthermore, the assessment of neurofilament light-chain in the cerebrospinal fluid (CSF) of another cohort of ALS and ALSci patients (Table 1; cohort 3) revealed comparable levels (ALS = 7950 ± 2030 pg/ml, ALSci = 11,068 ± 3073 pg/ml. 2-tailed unpaired t test; p = 0.39), suggesting the presence of similar neuronal breakdown across the brain in both groups (Supp. Fig. 6). When comparing synapse density in the frontal cortex to the motor cortex within the healthy control brains, we discovered a lower density in the motor cortex (Supp. Fig. 7; 2-tailed paired t test p = 0.036). This difference in regional synapse density was also observed in the larger group of cognitively unimpaired ALS cases (Supp. Fig. 7; 2-tailed paired t test p = 0.005), but was lost in the ALSci cases (Supp. Fig. 7; 2-tailed paired t test p = 0.75). This may suggest that synapse densities in the frontal cortex of ALSci brains had dropped to levels similar to those found in the motor cortex. We found no association between beta-amyloid, pTau or pTDP-43 pathology in BA4 and synapse density (Fig. 3g–i). Further, we found no association between beta-amyloid (Fig. 3j), pTau (Fig. 3k) or pTDP-43 (Fig. 3l) pathology in the motor cortex and cognitive score.

Glial cells are thought to play a prominent role in ALS pathogenesis [40], so we analyzed GFAP and CD68 burdens as a measure of astrocyte and microglial activity, respectively, in both BA9 and BA4. We found no correlation between GFAP burden and synapse density (Supp. Fig. 8) in BA9 (Pearson r = 0.127, p = 0.51) or in BA4 (Pearson r = 0.16, p = 0.44). There was also no correlation between CD68 burden and synapse density (Supp. Fig. 8) in BA9 (Pearson r = − 0.012, p = 0.95) or BA4 (Pearson r = 0.036, p = 0.86). Furthermore, there was no difference in GFAP or CD68 burden in BA9 (Supp. Fig. 8) in ALS patients compared to ALSci (2-tailed unpaired t tests. GFAP: p = 0.09; CD68: p = 0.59). This suggests that synapse loss in the ALSci patients is not associated with altered gliosis.

Taken together, these data suggest that synapse degeneration, occurring specifically within frontal (and not motor) cortex of ALS patients with cognitive deficits, is one underlying neurobiological contributor to cognitive decline. Of all of the pathological changes measured, synapse degeneration in BA9 was the strongest predictor of cognitive impairment.

We have optimized a combination of cognitive and genetic profiling with post-mortem tissue collection, neuropathology and high-resolution subcellular analyses. Our human tissue collection and processing protocol [26] robustly preserves synaptic structures for both electron microscopy and array tomography analysis. To the best of our knowledge, this is the only ALS cohort in the world that has been extensively characterized to this level of detail.

A few previous studies attempted to assess synaptic changes in human post-mortem tissue using less direct measurements such as optical density of immunohistochemical staining or western blotting [25, 45], but failed to observe any difference between ALS patients and controls. However, one potential confound is that we do not know the cognitive status of the patients in these studies [25, 45] and taken together our study highlights the importance of using highly precise, high-resolution approaches (such as array tomography and TEM) when studying synaptic changes in post-mortem material. One very recent study has detected a loss of synaptic terminals in the putamen and caudate of ALS patients, using immunohistochemical approaches [44]. Patients with severe cognitive decline meeting the diagnosis of FTD or FTD-MND had significant synaptic loss, whereas the ALS group had a more variable phenotype. Unfortunately, the ALS group was not stratified by cognitive status, so we cannot discern whether the synapse loss is related to cognitive decline in the ALS cases [44].

Synapse loss is the strongest correlate of cognitive decline in Alzheimer’s disease [54] and a recent meta-analysis highlighted presynaptic terminals as the major site of early pathogenesis in the disease [14]. Therefore, the loss of synapses from the frontal cortex of ALS patients described here fits well with the features of other disorders where cognitive decline is observed.

ALS mouse models, including the hSOD1^G93A and FUS^R521C models, have revealed pre-symptomatic spine loss in the motor cortex [18] and sensorimotor cortex [43], respectively. Intriguingly, in the hSOD1^G93A model, early pre-symptomatic loss of spines on pyramidal cells in the medial prefrontal cortex preceded changes in dendritic arborization [17]. The earliest onset of anatomical change occurred in the motor cortex, with such severe neuronal breakdown that cortical thinning was detected at pre-symptomatic time-points [17]. Thinning of the medial prefrontal cortex was only detected in adult diseased mice, by which time spine loss and dendritic changes were significant [17]. This temporal pattern of cortical breakdown observed in mouse models fits with our observations in human post-mortem tissue. We observed synapse loss in the prefrontal cortex without neuron loss, yet found significant cortical thinning in the ALS motor cortex. Furthermore, our CSF results suggest that similar levels of neuronal breakdown occurred across the brain of ALS and ALSci groups (Supp. Fig. 5). This is most likely due to the cortical thinning observed in the motor cortex and supports our finding that synapse loss in the frontal cortex of ALSci patients is not due to neuronal loss.

Two human post-mortem studies have found greater expression of TDP-43 aggregates in the frontal cortex of patients with cognitive decline, compared to cognitively unimpaired ALS patients [9, 13]. TDP-43 aggregates are often found in the brain of ALS patients, and a sequential staging of pathology progression through the central nervous system has been described [8]. It is believed that TDP-43 pathology may progress through synaptically connected brain regions [6, 16], in a similar manner to other neurodegenerative diseases [7], thus placing synapses at a crucial point in ALS pathogenesis. Furthermore, in a human TDP-43 (hTDP43) overexpressing mouse model, synaptophysin loss was noted in whole brain homogenates [33] suggesting that altered TDP-43 function may drive synapse loss. Intriguingly, four studies have described what appears to be “synaptic” labeling of pTDP-43 in the neuropil of human post-mortem cortex from ALS and FTD patients [4, 37, 44, 53]. A recent study described pTDP-43 puncta in very close proximity to synaptophysin puncta in human ALS brain [53], with the suggestion that small pTDP-43 aggregates may be accumulating in dendritic spines. However, another recent study suggests pTDP-43 puncta co-localize with synaptophysin [44], thus suggesting these small aggregates are in fact presynaptic. In support of this, we also noted punctate labeling of pTDP-43 in the neuropil of some of our ALS cases and found pTDP-43 puncta in a number of synapses in a subset of ALS cases (Fig. 1i, j). Given that synaptic accumulation of pathological proteins in Alzheimer’s disease and FTLD-tau animal models drives synapse loss [27‐29, 51, 52], it is possible that a similar build-up of pTDP-43 in the ALS synapse may initiate synapse breakdown. An alternative explanation is the phagocytosis of synaptic components by pathologically active glial cells, as recently described in Alzheimer’s disease [23]. In support of this, we recently observed that when TDP-43 is specifically knocked out in microglial cells, they adopt a hyper-phagocytic phenotype. The microglial cells actively phagocytozed synapses, resulting in a significantly lower synaptic density in the cortex of these transgenic mice compared to controls [39]. Furthermore, in a human cohort of ALS patients, pTDP-43 positive cases had significantly higher expression of the phagocytic microglial marker CD-68 [39]. Interestingly, in our current study, we found the same result in the frontal cortex of ALS patients. Those with pTDP-43 in the frontal cortex had a higher CD-68 burden than ALS cases without pTDP-43 (pTDP-43-positive n = 14, pTDP-43-negative n = 20; 2-tailed unpaired t test; p = 0.028). However, CD68 expression did not correlate with synaptic density (Pearson r = − 0.012; p = 0.95) and thus further work is required to understand the role of phagocytic microglia in synapse loss.

We found no direct association between the presence of the dementia-associated pathologies, beta-amyloid and pTau in the frontal cortex and lower ECAS score (Fig. 2). This is in agreement with recent studies showing no association between Alzheimer’s pathology and cognition in ALS [9, 42]. However, it is interesting to note that 5 out of the 7 ALSci cases were positive for pTau and 5 of the 7 pTau + ve cases were ALSci. This finding aligns with the proposed model that certain forms of pTau in the frontal cortex may associate with cognitive decline in ALS [50, 59, 60] and warrants further study once more cases become available to us. One of the limitations of this study is the small number of cognitively impaired patients who have donated brain tissue for these synaptic analyses. As more people donate tissue, we may be able to further dissect the contributions of pathological protein accumulation within synapses and genetic contributors to synaptic loss and cognitive decline. Another potential confound in the study is that the control group prepared for array tomography imaging was significantly older than the ALS group (Table 1). However, since subtle age-related changes in synapse loss have been reported [36], we predict that this age discrepancy would make the detection of synapse loss in ALS more difficult and thus are confident that this age difference did not lessen the main finding of synapse loss in frontal cortex of ALS cases, which is worse with cognitive impairment. In support of this idea, we did not detect any correlation between age and synapse density in our data (Pearson r = 0.23, p = 0.23).

In summary, we have combined two high-resolution imaging techniques to reveal synapse loss in the frontal cortex of ALS patients. We show that ALS patients with cognitive impairment have lower synaptic densities than controls. Synapse density did not differ in the primary motor cortex between ALS and controls, most likely due to the significant cortical thinning we observed in the ALS motor cortex. These findings significantly advance our understanding of ALS-associated neuropathology and identify the underlying neurobiological substrate for cognitive decline in ALS. Further, these results suggest potential synaptic biomarkers for predicting which ALS patients will experience cognitive changes, and point to the potential of synapse-directed therapeutics to combat both the cognitive and potentially also the motor symptoms of ALS.