Introduction

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by the involvement of upper and lower motor neurons1,2. The impairment of upper motor neurons (UMNs) primarily affects the corticospinal tract (CST) and corpus callosum (CC)3,4, whereas lower motor neuron involvement primarily affects spinal anterior horn cells and their peripheral nerves5,6,7,8,9. However, studies have also confirmed various degrees of CC involvement in different ALS subtypes10. Both cross-sectional and longitudinal studies indicate that neuroimaging biomarkers have value in differentiating and predicting ALS11,12.

White matter microstructural differences and abnormalities can be captured in vivo via diffusion magnetic resonance imaging (dMRI)13, with diffusion tensor imaging (DTI) being the primary technology used. Each fiber tract's different DTI parameters are considered a neuroimaging phenotype, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), radial diffusivity (RD)14, and mode of anisotropy (MO)15,16. Among these, FA serves as the primary metric of interest in many studies; it is a robust measure of overall integrity and directionality and is highly sensitive to general connectivity changes17. Previous studies have reported abnormalities in DTI, such as a decrease in FA in the white matter fibers of ALS patients. The MD, AD, and RD directly quantify the absolute magnitude of directionalities. They are more sensitive to specific types of microstructural changes18. Additionally, MO can characterize the anisotropy type and describe whether the shape of the diffusion tensor is more linear or planar15. Longitudinal studies have demonstrated that the progression of ALS leads to dynamic changes in neuroimaging markers3,11. Although the development of ALS involves specific fiber tracts, suggesting causality, whether the phenotype of white matter fibers can influence ALS has not been confirmed in relevant studies.

We used a bidirectional two-sample Mendelian randomization (TSMR) approach19, which examines causation by exploiting strong genetic variants related to neuroimaging phenotypes or ALS as instrumental variables (IVs), to explore the associations of the genetic architecture of white matter fiber tracts with the clinical events of ALS. Moreover, multivariable Mendelian randomization (MVMR), which can explain whether an association between different exposures and outcomes is mediated by other factors, was performed to determine the pathophysiological mechanisms underlying the associations20. Since genetic variants are randomly allocated at conception, they are mostly unaffected by potential confounding factors, which can introduce bias in the associations being studied. Additionally, genetic variants are not influenced by acquired predispositions, which helps prevent bias from reverse causation. Therefore, Mendelian randomization (MR) is well suited for investigating potential associative relationships between neuroimaging phenotypes and the risk of developing ALS.

Methods

Study design

As shown in Fig. 1, we used available summary results from genome-wide association studies (GWASs) to perform MR and examine the associative relationships between neuroimaging phenotypes and ALS. The design of our study included three steps: (1) the acquisition of summary-level GWAS data for neuroimaging phenotypes and ALS; (2) the identification of genetic variants to serve as IVs; and (3) the estimation of the effects of neuroimaging phenotypes on ALS. The theoretical basis of MR research relies on three assumptions: assumption 1, the selected genetic variants are not related to other confounders; assumption 2, the selected genetic variants are significantly related to the exposure; and assumption 3, the selected genetic variants are significantly related to the risk of outcome only through the exposure21.

Figure 1
figure 1

Flow chart showing the process for the MR analyses. (A) The TSMR analysis was conducted to assess the association between specific neuroimaging phenotypes and the risk of ALS, as well as to fulfill assumptions 1 to 3. (B) The MVMR analysis was used to investigate the roles of various neuroimaging metrics within specific fiber tracts or the average values of the 21 white matter fibers in ALS. MR, Mendelian randomization; TSMR, two-sample Mendelian randomization; MVMR, multivariable Mendelian randomization; IVs, instrumental variables; GWAS, genome-wide association study; CST, corticospinal tract; GCC, genu of the corpus callosum; BCC, body of the corpus callosum; SCC, splenium of the corpus callosum; dMRI, diffusion magnetic resonance imaging; FA, fractional anisotropy; AD, axial diffusivity; MD, mean diffusivity; RD, radial diffusivity; MO, mode of anisotropy; PC1, the first principal component of fractional anisotropy; PC5, the fifth principal component of fractional anisotropy; IVW, inverse variance weighting method; SNPs, single-nucleotide polymorphisms.

Full size image

Genetic associations with neuroimaging phenotypes

The GWAS dataset22 involved 33,292 European individuals and contained individual-level DTI data from five data sources: the UK Biobank (UKB); the Adolescent Brain Cognitive Development (ABCD) study; the Human Connectome Project (HCP); the Pediatric Imaging, Neurocognition, and Genetics (PING) study; and the Philadelphia Neurodevelopmental Cohort (PNC). The images from these sources were harmonized via the ENIGMA-DTI pipeline (http://enigma.ini.usc.edu/protocols/dti-protocols/), which produced voxelwise DTI maps for all the subjects. DTI maps were utilized to construct 21 predefined cerebral white matter tracts. Each tract was evaluated using the FA, MD, RD, AD, MO, and the top five principal component scores of FA were determined via functional principal component analysis (FPCA). Due to the selective involvement of motor fibers in ALS and the frequent occurrence of CC involvement in ALS neuroimaging studies, we selected four fiber tracts for our study: the CST3,4, genu of the CC (GCC)10,23,24, body of the CC (BCC)3,4, and splenium of the CC (SCC)25,26. Additionally, we employed the five imaging metrics mentioned above to represent the global average across 21 fiber tracts and incorporated them into the MR analysis.

Genetic associations with amyotrophic lateral sclerosis

The publicly available European-based GWAS summary statistics for ALS included more than 10 million single-nucleotide polymorphisms (SNPs) in 27,205 ALS patients and 110,881 controls27. Patients were not selected based on a family history of ALS; they were diagnosed and ascertained through specialized motor neuron disease (MND) clinics where they were diagnosed with ALS according to the (revised) El Escorial Criteria by neurologists specializing in MNDs. This dataset comprises 10,461,755 variants with a minor allele frequency (MAF) as low as 0.1% according to the Haplotype Reference Consortium resource. All detailed information (e.g., study subjects, genotyping, quality control, and statistical analyses) and data availability statements can be found in the original article27.

IVs selection

We used the same criteria for the bidirectional two-sample MR analysis when we selected the IVs. SNPs that were independently (r2 < 0.001) associated with each exposure at the genome-wide significance level (P < 5 × 10−8) and had an MAF > 0.001 were selected as IVs. SNPs that could not be found in the outcome data were replaced with proxy SNPs in strong linkage disequilibrium (LD) (r2 > 0.9) by searching the LDlink website (http://ldlink.nci.nih.gov/). If a proxy SNP was not reported, the SNP was excluded from downstream MR analyses. Due to the requirements of MR approaches, an exposure was removed when there were fewer than two available SNPs. SNPs that were correlated with the outcome (P < 0.05) were excluded.

Statistical analysis

In the primary analysis, the fixed-effect inverse variance weighted (IVW) method was used to calculate the overall effects in the absence of heterogeneity28. However, if heterogeneity was observed between the causal estimates of genetic variants, a multiplicative random effects (MRE) IVW model was conducted. Through forward MR analysis and MVMR, we inferred the associative relationships of neuroimaging phenotypes on ALS by reporting the odds ratio (OR) for genetic variants. In the reverse MR analysis, we inferred the associative relationships of ALS on neuroimaging phenotypes by reporting the β value, which represents the standard deviation of changes in imaging-derived phenotypes caused by ALS.

Sensitivity analysis

For the sensitivity analyses, we utilized the weighted median, simple median29, and MR‒Egger methods30. MR‒Egger regression, which accounts for the presence of pleiotropy when the intercept significantly deviates from the origin, and MR-PRESSO analysis, which can identify SNP outliers and provide results identical to those of IVW after removal of outliers, were used to evaluate the potential pleiotropy of IVs31. The heterogeneity of the SNPs utilized in the IVW estimates was tested with Cochran’s Q test, and P values less than 0.05 indicated the presence of heterogeneity. Leave-one-out analysis was employed to evaluate the possibility of results being driven by a single SNP. Additionally, we calculated F statistics for the IVs to determine whether they were strong instruments. Given the close correlation between different neuroimaging metrics, an MVMR approach was employed to reduce mutual interference. Specifically, within the CST, the associative relationship of FA with ALS was estimated with the other four metrics (AD, MD, RD, and MO) as covariates. The same MVMR approach was applied to the other fiber tracts. All analyses were conducted using the ‘TwoSampleMR’ package (version 0.5.7) in R version 3.6.3 (R Foundation for Statistic Computing, Vienna, Austria) (http://www.r-project.org/).32 A P value less than 0.05 after false discovery rate (FDR) correction was considered to indicate a valid positive result. An unadjusted P value less than 0.05 was regarded as a suggestive indicator. The process is shown in Fig. 1.

Ethical issues

Additional ethical approval was not required since this study was based on MR analysis and depended on summary-level GWAS data rather than individual-level data, and the databases used in this analysis have been published or shared previously.

Results

Forward MR: the putative associative relationships of neuroimaging phenotypes on ALS

In our study, proxies related to well-known neuroimaging phenotypes were utilized to investigate the relationships between neuroimaging phenotypes and ALS using various MR methods. Altogether, 519 SNPs were identified as IVs for neuroimaging phenotypes.

We found that an increased GCC PC1 had a significant effect on the risk of ALS (Fig. 2). The IVW odds ratio (OR) for GCC PC1 was 1.363 (95% confidence interval (CI) 1.178–1.577, PFDR = 0.001). We also found that increased CST MD, CST RD, GCC FA, and BCC PC1 had potential effects on the risk of ALS, although these phenotypes were not corrected for multiple comparisons. Specifically, for CST MD, the IVW OR was 1.214 (95% CI 1.014–1.454, Puncorrected = 0.035, PFDR = 0.194); for CST RD, the OR was 1.220 (95% CI 1.047–1.422, Puncorrected = 0.011, PFDR = 0.143); for GCC FA, the OR was 1.234 (95% CI 1.060–1.436, Puncorrected = 0.007, PFDR = 0.130); and for BCC PC1, the OR was 1.205 (95% CI 1.020–1.423, Puncorrected = 0.028, PFDR = 0.194) (Supplementary Data S1). When the MR‒Egger method was used, GCC PC1 generated similar results (OR = 2.683, 95% CI 1.583–4.550, PFDR = 0.029). The MR‒Egger method revealed no evidence of associations among CST MD, CST RD, GCC FA, and BCC PC1 (PFDR = 0.801, 0.756, 0.112 and 0.756, respectively). However, the direction of the effect was consistent between MR‒Egger and IVW. Estimates based on the simple median method and weighted median method showed similar trends.

Figure 2
figure 2

Odds ratios for the associations between different genetically predicted neuroimaging phenotypes and ALS. The main results for the effects of a 1-unit increase in genetically predicted parameter values on ALS using MR approaches. The P value is an uncorrected value. OR, odds ratio; CI, confidence interval; MR, Mendelian randomization; IVW-MRE, inverse variance weighted Mendelian randomization estimation; SNPs, single-nucleotide polymorphisms; CST, corticospinal tract; MD, mean diffusivity; RD, radial diffusivity; GCC, genu of the corpus callosum; FA, fractional anisotropy; PC1, first principal component of fractional anisotropy; BCC, body of the corpus callosum.

Full size image

The MR–Egger intercept analysis did not indicate horizontal pleiotropy (all PFDR values > 0.05). However, the MR-PRESSO global test suggested horizontal pleiotropy for GCC PC1 (PFDR = 0.016). Fortunately, the MR-PRESSO distortion test results for the GCC PC1 were not significantly different. The heterogeneity of the GCC FA and GCC PC1, which may increase the probability of type I errors, was indicated by Cochran’s Q test (GCC FA: Q = 40.693, PFDR = 0.027; GCC PC1: Q = 38.146, PFDR = 0.042) (the detailed sensitivity analysis can be found in Supplementary Data S2-4). Therefore, we applied the IVW-MRE model for the MR analysis instead of the fixed-effects IVW model (Fig. 2).

On the other hand, a greater CST PC1 and GCC RD are potentially associated with a reduced risk of ALS (Fig. 2). The IVW analysis results indicated that the risk of ALS decreased by 17.2% and 15.4% with a genetically predicted one standard deviation (1-SD) increase in CST PC1 (OR = 0.828, 95% CI 0.697–0.983, Puncorrected = 0.031, PFDR = 0.194) and GCC RD (OR = 0.846, 95% CI 0.728–0.984, Puncorrected = 0.030, PFDR = 0.194). This potential association was confirmed by the MR‒Egger method for GCC RD (OR = 0.363, 95% CI 0.219–0.605, PFDR = 0.028), whereas CST PC1 showed a similar trend, although it was not statistically significant (OR = 0.744, 95% CI 0.206–2.688, PFDR = 0.801). Estimates from the weighted median method and simple median method showed similar trends but did not confirm associations significantly. The MR‒Egger intercept revealed no evidence of horizontal pleiotropy for CST PC1 and GCC RD (all PFDR values > 0.05). Similar results were obtained with the MR-PRESSO global test (all PFDR values > 0.05). Cochran’s Q test indicated that GCC RD exhibited heterogeneity (Q = 56.509, PFDR = 0.003), whereas CST PC1 showed no heterogeneity (Q = 1.353, PFDR = 0.755). Therefore, we also employed the IVW-MRE method for analysis (Fig. 2).

Scatter plots were generated to estimate the effects of various neuroimaging phenotypes on ALS through each SNP (Fig. 3). Figure 4 summarizes the potential associative relationships described above.

Figure 3
figure 3

Scatterplot of the potential effects of SNPs on different imaging phenotypes and ALS. The slope of each line corresponds to the estimated MR effect of each method. The phenotypes included (A) CST MD, (B) CST RD, (C) GCC FA, (D) GCC PC1, (E) BCC PC1, (F) CST PC1 and (G) GCC RD. SNP, single-nucleotide polymorphism; MR, Mendelian randomization; ALS, amyotrophic lateral sclerosis; CST, corticospinal tract; MD, mean diffusivity; RD, radial diffusivity; GCC, genu of the corpus callosum; FA, fractional anisotropy; PC1, first principal component of fractional anisotropy; BCC, body of the corpus callosum.

Full size image
Figure 4
figure 4

Potential relationships between changes in the neuroimaging metrics of the CST, GCC, and BCC and the risk of ALS. The asterisk indicates that the P value of this neuroimaging phenotype passed the FDR correction. ALS, amyotrophic lateral sclerosis; CST, corticospinal tract; MD, mean diffusivity; RD, radial diffusivity; GCC, genu of the corpus callosum; FA, fractional anisotropy; PC1, first principal component of fractional anisotropy; BCC, body of the corpus callosum.

Full size image

Reverse MR: the putative associative relationships of ALS on a specific neuroimaging phenotype

In the reverse MR analysis, 11 SNPs were identified as IVs for ALS. As shown in Supplementary Data S5, we did not detect significant associations of the risk of ALS on specific neuroimaging phenotypes.

Multivariable MR analyses

We conducted separate MVMR analyses using five neuroimaging phenotype indicators (FA, AD, MD, RD, and MO) as exposures for each fiber tract (CST, GCC, BCC, and SCC) and the average values of 21 fiber tracts. As shown in Fig. 5, the results indicate that within any of the fiber tracts, each neuroimaging metric had no direct impact on the risk of ALS after adjustment for the other four neuroimaging metrics.

Figure 5
figure 5

Odds ratios and 95% CIs for the effects of each fiber tract (CST, GCC, BCC, and SCC) and the average values of 21 fiber tracts on ALS estimated using the MVMR IVW approach.

Full size image

Relationship between selective SNPs and ALS–frontotemporal lobar degeneration (FTLD)-related genes

Due to the presence of pathogenic mutations in 5%-10% of ALS patients33 and the relatively prominent changes in neuroimaging phenotypes in ALS–frontotemporal lobar degeneration (FTLD) patients34,35, we matched SNPs with potential associative relationships in this MR analysis with several ALS–FTLD-related pathogenic genes to identify specific ‘gene–neuroimaging phenotype‒ALS’ associations; the main genes included C9orf72, CHCHD10, SQSTM1, TBK1, CCNF, FUS, TARDBP, OPTN, UBQLN2, TURA4A, ATXN2, VCP, and CHMP2B36. Unfortunately, no known genes are definitively located within SNPs associated with neuroimaging phenotypes.

Discussion

Both cross-sectional and longitudinal neuroimaging studies have confirmed that, compared with normal controls, ALS patients exhibit changes primarily in the CST and CC in the early stages of the disease, with a notable prevalence of UMN involvement26,37. Similar findings have also been reported in asymptomatic individuals carrying pathogenic gene variants38. However, whether the baseline condition of white matter fiber tracts contributes to the onset of clinical events in individuals with ALS and whether ALS can cause damage to specific fiber tracts remain unclear. Currently, dMRI is the optimal noninvasive method for detecting in vivo changes in white matter fiber tracts13,39.

Both FA and RD are valuable indices in DTI studies because they provide insights into the structural integrity of white matter tracts40. Higher FA values indicate greater directionality and organization of fibers, suggesting healthier and more intact white matter tracts. Increased RD values suggest increased diffusion in directions perpendicular to the fibers, which can be indicative of myelin degradation or loss. FA and RD can be used to detect subtle changes or abnormalities in patients with various neurological conditions, including ALS41,42. Additionally, MD is commonly used to assess and study the microstructure and integrity of the brain and neural tissues43. Higher MD values are often associated with pathological changes in the tissue. In DTI models, the FPCA is utilized to capture the most prominent variations in components of FA within each tract. This approach is designed to provide a deeper understanding of axonal organization and myelination, which are often overlooked when tract-averaged values are used44. These principal components (PCs) may represent FA changes that are more relevant to specific clinical outcomes. In summary, in previous ALS neuroimaging studies, higher FA and PC values and lower MD and RD values were associated with better fiber tract integrity, whereas the opposite results indicated poorer integrity.

Using data from large-scale neuroimaging phenotype GWASs and an ALS GWAS, we conducted a bidirectional two-sample MR study and reported an association between GCC PC1 and the risk of ALS. Additionally, although CST MD, CST RD, GCC FA, GCC PC1, BCC PC1, and GCC RD did not pass the corrected P value threshold in the IVW method, they still showed some genetically predicted significance. Overall, these findings elucidate the importance of the CST and commissural fibers in the mid-anterior region in ALS pathogenesis.

Since multiple parameters are available to reflect the integrity of fiber tracts, our results of the forward MR analysis indicate that, in the uncorrected P value situation, changes in the integrity of each fiber tract (CST, GCC, and BCC), as measured by different DTI metrics, are consistent with the risk of ALS. For example, as depicted in Fig. 2, higher RD and MD in the CST or lower FA indicate poorer CST integrity, potentially increasing the risk of ALS. Conversely, for the CC fiber tract, the relationship is the opposite. Higher FA and PC1 in the GCC, higher PC1 in the BCC, or lower RD in the GCC suggest better integrity in the mid-anterior CC, potentially increasing the risk of ALS.

The forward MR analysis suggested that as the integrity of the GCC increases, the risk of ALS increases. The current understanding suggests that when the integrity of commissural fibers (mainly the CC) is better, the functional connectivity between the two cerebral hemispheres is stronger and more reliable45,46,47,48. Since ALS affects these white matter fibers, we hypothesize that damage to these fibers leads to decreased integrity, resulting in a loss of the structural basis for functional connections and promoting ALS development. In ALS patients, the possibility of pathological protein propagation, such as TDP-43 proteinopathy (which is currently the most significant), is important to consider49,50. However, pathological studies of ALS have not revealed TDP-43 protein deposits in the CC51. Instead, they mainly show reduced silver-stained fibers, along with GFAP-positive glial cells and CD68-positive cell infiltrates52,53. The following question remains: How does the integrity of the GCC affect the risk of ALS? ALS patients initially present with focal symptoms, and pathological changes in ALS patients are related to the onset site, resulting in focal characteristics. More importantly, the excitotoxicity of glutamate in the motor cortex plays a particularly important role in ALS pathogenesis54. Based on the results of this study, we speculate that the better integrity of the GCC may facilitate the spread of pathogenic proteins to both brain hemispheres. Therefore, in the presymptomatic state, when excitotoxic damage occurs on one side of the motor cortex (assumed to be M1), the information of excitability is transmitted along the CC fibers to the homologous cortex on the opposite side (assumed to be M1'). Since the CC has mutual homotopic callosal inhibition55, M1' receives inhibitory information, leading to a decrease in the excitability of the M1' cortex. However, the amount of excitatory information transmitted from M1' to M1 via the CC increases (due to mutual homotopic callosal inhibition), enhancing the excitotoxic effect on the M1 cortex. After repeated cycles, when the excitotoxic effect of the M1 cortex reaches the pathogenicity threshold, clinical symptoms appear. The integrity of the CC plays a key role in this information transmission process. The increased integrity of the CC facilitates efficient interhemispheric information transmission, thereby accelerating the spread of pathological information and increasing the risk of ALS. Conversely, the compromised integrity of the CC slows the transmission of pathological information, thus reducing the risk of ALS (Fig. 6). Nevertheless, further research that combines analyses of brain functional connectivity with DTI is essential for confirmation. As part of the commissural fibers, although the DTI parameters associated with the BCC did not pass multiple corrections, their potential research significance suggested by the MR results is consistent with that of the GCC. Therefore, the effects of the GCC and BCC on ALS should be taken seriously.

Figure 6
figure 6

Speculated mechanism of the ALS risk related to the CC. (A) The pathway of excitotoxic information transmission from the M1 cortex through the CC to the M1' cortex and the feedback loop. This pathway may be influenced by the integrity of the CC, potentially affecting the risk of ALS. (B) corresponds to a–c in (A), where red represents excitatory stimuli or information and blue represents inhibitory stimuli or information. (C) corresponds to b–d in (A), and the color representation is the same as that in (B). CC, corpus callosum.

Full size image

On the other hand, we also need to pay attention to the potential significance of the CST. When the integrity of the CST is better, the risk of ALS is somewhat reduced. We believe that the CST is the primary affected fiber tract in ALS patients and is closely associated with the pathology of cortical motor neurons56,57,58. When its integrity is compromised to a specific threshold, the onset of ALS occurs; therefore, fiber tracts with better integrity result in a slower onset of the disease59. This finding is supported by existing research3 indicating that the integrity of the CST decreases as the severity of ALS pathology increases, for clearer expression.

Our study included an assessment of the global average values of 21 fiber tracts for 5 neuroimaging phenotypes to avoid excessive subjectivity in our exposure selection. The results confirmed that the exposures had no impact on ALS. Microstructural differences and changes may not exhibit a consistent pattern across all fiber tracts, and ALS may affect specific fiber tracts, which could be neutralized when subjected to a comprehensive average assessment60.

Conversely, in the reverse MR analysis, we failed to find strong evidence that a genetic predisposition for ALS was associated with specific fiber tracts. This result does not support the abovementioned research background. This discrepancy may be because ALS primarily affects motor neurons (Betz cells in the motor cortex and motor neurons in the anterior horn of the spinal cord), while white matter fibers are secondary damaged structures, which cannot directly show that ALS could genetically predict the integrity of fiber tracts. Additionally, DTI is not the only imaging technique used to describe white matter fiber structures; other methods, such as neurite orientation dispersion and density imaging (NODDI)61, can also be used to evaluate fiber structures from different perspectives.

Some SNPs overlapped among the IVs for selected neuroimaging phenotypes, which may affect the results; therefore, the MVMR approach, sensitivity analyses, and horizontal pleiotropy tests were adopted to assess the true relationship and detect the robustness of the estimates. However, according to the MVMR analysis, no significant positive results were found for either specific fiber tracts or the average values across all brain fiber tracts. This result may be due to a mutual constraint relationship among the DTI metrics within specific fiber tracts62, making pinpointing risk factors further by controlling a single parameter challenging.

The exact underlying mechanism linking neuroimaging phenotypes to ALS remains elusive. Therefore, in conjunction with the results of our data analysis, we included pathogenic genes related to ALS–FTLD to determine whether any overlap occurred with SNPs showing positive results. However, no overlap was found. In the future, the impact of genes on neuroimaging phenotypes should be further explored.

Our study is the first to investigate the effects of neuroimaging phenotypes on ALS patients. All the analyses were performed using data from the largest European-based GWASs. Nevertheless, some shortcomings still exist in this study. The data from the MR analysis indicate heterogeneity and horizontal pleiotropy in some exposures, particularly in the GCC tract. Therefore, we switched from the IVW method to a random effects model for analysis. Additionally, using MR-PRESSO, we found that after removing outliers, the significance remained unchanged, confirming the reliability of the results. Furthermore, because each neuroimaging phenotype for each fiber tract was treated as a separate IV, this MR analysis included many IVs. After FDR correction, only GCC PC1 still showed significant differences, suggesting the involvement of other tracts, such as the BCC and CST, in the risk of ALS. Additionally, the FDR assumes that all the statistical tests have a similar ability to detect potential discoveries. However, FDR estimation is subject to variability due to differences in the underlying biology, signal-to-noise ratio, or features of the trait, which can lead to greater power than other methods in certain tests. Although we included 45 neuroimaging phenotypes as instrumental variables, this number is still far from sufficient. The study investigated only specific fiber tracts via DTI and did not maximize the expansion to all brain fiber tracts. Our failure to identify specific neuroimaging phenotypes associated with ALS in the reversed MR analysis, suggesting the need for larger datasets or additional assessment methods to characterize such associations.

More clinical studies and animal experiments are needed to verify the preliminary results of this study. The ‘gold standard’ for empirically alleviating the concerns of residual confounding and reverse causation in clinical research is an RCT. However, RCTs testing the associations between neuroimaging phenotypes and the risk of ALS have not been performed, mainly because of the challenges of implementation and the time-consuming nature of these studies. The presymptomatic stage of ALS can persist for years, and cohort studies may provide more credible evidence with a diminished interference of reverse causation than other observational studies. A large-scale population-based prospective cohort can be established, and baseline information can be collected through head magnetic resonance imaging (MRI) scans and neurological examinations. With long-term follow-up, the relationships between the observed neuroimaging phenotypes and ALS can be revealed using various association analyses. Combining these data and the results generated with the MR framework may yield convincing conclusions in the future. Although our MR analysis cannot fully substitute for randomized controlled trials evaluating intervention effects, it provides a guide for the design of future costly experiments.

Conclusions

This study provides genetic evidence of potential associative relationships between neuroimaging phenotypes and ALS. The results suggested that increased CST MD, CST RD, GCC FA, GCC PC1, and BCC PC1 are potential risk factors for ALS, whereas increased CST PC1 and GCC RD have a protective effect, possibly due to the functional characteristics of different fiber tracts. The genetically predicted results suggest that better fiber integrity in the GCC and BCC is associated with an increased risk of ALS, whereas better fiber integrity in the CST, which is directly linked to cortical motor neurons, is associated with a lower ALS risk. These findings cannot be explained by currently identified ALS–FTLD-causative genes. Additionally, these results have potential implications for clinical practice, including the exploration of interventions such as antipsychotic medications and transcranial magnetic stimulation for ALS patients. However, validation through RCTs is needed to confirm the reliability of these findings.