Supplemental
Methods
ALS sample
Penn UMN score
The Penn UMN score ranged from 0 to 32 points and comprised items from the bulbar
segment (0-4 points) and from each of the four limbs (0-7 points per limb) [1]. In detail, for
the bulbar segment, single points were allocated for an abnormal jaw-jerk reflex, an
abnormal facial reflex, the existence of the palmomental sign and the existence of an
abnormal pseudobulbar affect. For the upper extremity subscore, single points were given for
each, pathologically brisk biceps reflex, triceps reflex, presence of finger flexors, Hoffmann’s
sign or the existence of a clonus anywhere in the limb. Additionally, spasticity was rated
according to the Ashworth Spasticity Scale (0-2 points, with adding 0 points for Ashworth 1
(normal tone), 1 point for Ashworth 2-3, 2 points for Ashworth 4-5) [2]. For the lower
extremity subscore, single points were allocated for each, pathologically brisk plantar reflex,
ancle reflex, crossed adduction, Babinski’s sign and a clonus anywhere in the limb. For the
lower extremity, spasticity was rated the same way as described for the upper extremity
subscore.
Clinical phenotypes
Clinical phenotypes were classified according to recent specifications [3, 4]. At the time of
study inclusion, a variable combination of UMN signs (spastic tone, clonus, etc.) and LMN
signs (wasting, weakness, fasciculations) in the upper and lower limbs were found in those
designated as classic ALS who, in turn, fulfilled the El Escorial criteria of definite or probable
ALS. UMND ALS patients had either no LMN signs, or, if present (1) they were restricted to
only 1 neuraxis level (bulbar, cervical, or lumbosacral); and (2) electromyographic
abnormalities were limited to sparse fibrillation potentials/positive sharp waves or minor
enlargement of motor unit potentials in 1 or at most 2 muscles [5, 6] for at least 12 months
after symptom onset. The diagnostic criteria for PLS required a period of at least 4 years in
which there were only UMN signs on examination. Other conditions that mimic PLS, such as
hereditary spastic paraplegia (HSP) were excluded by appropriate investigations [7]. All
patients with LMND ALS had clinical and electrophysiological evidence of sporadic
progressive pure LMN involvement in 1 or more regions without clinical signs of UMN
dysfunction. To differentiate this condition from early limb-onset ALS, we specified that LMN
involvement must be the predominant finding for at least 12 months after the symptom onset.
LMND ALS comprised patients with flail arm phenotype (n=4), flail leg phenotype (n=2) and
progressive muscular atrophy (n=3). Other LMN diseases, such as multifocal motor
neuropathy, spinal muscular atrophy, monomelic amyotrophy, Kennedy’s disease, and post-
polio syndrome, were excluded by extensive clinical and laboratory examinations [7, 8].
Data availability
CSF data were on hand for all ALS patients, of those 89 cases, 58 (69%) and 13 (15%)
patients, respectively, have already been included in our previous cross-sectional and
longitudinal peripheral nerve sonography ALS studies [3, 9, 10]. Out of the 84 patients with
available baseline ALSFRS-R scores, longitudinal ALSFRS-R scoring was performed in n=71
cases (80%) with at least two follow-ups and n=46 cases (52%) with at least three follow-ups.
Survival data could be identified in n=86 subjects (97%) with n=53 (62%) having died after a
median survival time of 35.8 months. C9orf72 and SOD1 status was available in n=64
patients (72%), comprising n=6 (9%) suffering from familial ALS (n=2 with C9orf72 positivity
and n=4 with SOD1 positivity). Nerve CSA was available in n=72 (81%) cases, CMAP
amplitudes in n=65 (73%) and MPRAGE images in n=61 (69%) subjects of whom n=51
(57%) had also cerebral DTI measures. Constellations of individual data availability in ALS
are indicated in Supplemental Figure 1.
CSF measures
Within 20 minutes of lumbar puncture, CSF samples were centrifuged at 4 C, aliquoted and
stored at -80 C until analysis. CSF biomarkers were measured with commercially available
ELISA (for NfL: NF-light® ELISA, IBL International GmbH, Hamburg, Germany; for total tau
[ttau] or ptau: Innotest hTauAg or Innotest p-Tau, Innogenetics, Ghent, Belgium), following
the instructions provided by the manufacturer.
To assess the performance of the NfL assay we determined the intra-assay coefficient of
variability (CV; =reproducibility, within-assay performance) and the inter-assay CV
(=repeatability, between-assay performance) [11]. CV was calculated using the root mean
square method, described e.g. in [19]. CSF samples of 2 controls and four ALS patients were
measured twice on the first assay, and procedure was repeated 24 hours later taking a
second assay. Intra-assay CV of duplicates was 3.1%, inter-assay CV was 10.6%, which is
in line with the literature [11]. Detailed CSF NfL values of each sample are given in
Supplemental Table 1.
3T MRI measures of the brain
All MRI scans were performed on the same Siemens Verio 3 T system (Siemens Medical
Systems, Erlangen, Germany) with a 32-channel head coil. 3D MPRAGE images were
acquired using the following parameters: acquisition time 9 min, 20 s, repetition time 2500
ms, echo time 4.82 ms, inversion time 1100 ms, flip angle 7 °, voxel size = 1×1×1 mm
3
. DWI
data were acquired with a resolution of 2×2×2 mm
3
. Diffusion gradients were applied along
30 non-collinear directions with b = 1000 s/mm
2
, one scan without diffusion weighting (b =
0 s/mm
2
) was also acquired. The data were averaged across two repetitions (for full details
see [12, 13]). A T2-weighted FLASH sequence was acquired during the same session to
investigate the presence of white matter hyperintensities.
Diffusion tensor imaging analysis
Diffusion tensor images were processed using the FMRIB software library (FSL [14];
Analysis Group, FMRIB, University of Oxford, UK). In brief, each diffusion weighted volume
was affined-aligned to its corresponding b0 image using FSL's linear image co-registration
tool (FLIRT v5.4.2) to correct for motion artifacts and eddy-current distortions. Using FSL’s
brain-extraction tool (BET v2.1) a binary brain mask of each b0 image was generated, with
fractional threshold f = 0.1 and vertical gradient g = 0. The original b-matrix was reoriented
using an in-house script to adjust it for rotations induced by the previous transformations.
FSL's diffusion toolbox (FDT v2.0) was used to fit a single tensor model, taking a weighted
linear approach, and to compute the maps of DTI scalars (FA, mean diffusivity (MD), radial
diffusivity (RD), axial diffusivity (AD)). Load of white matter lesion was evaluated on a T2-
weighted FLASH sequence employing the Fazekas scale [15].
The analyses were performed employing tract-based spatial statistics [16] that warped all the
FA images to the FMRIB58_FA standard template (FMRIB; resolution: 1×1×1 mm
3
) in
MNI152 space using FSL's non-linear registration tool (FNIRT v1.0). The warped FA maps
were averaged to create a mean FA template, from which the FA skeleton was computed,
imposing an FA threshold of 0.2. All the FA maps as well as the maps of the other DTI
scalars were then projected onto the skeleton. The whole-brain regression analysis was
conducted employing the Randomise tool version 2.9 available in FSL with 5000
permutations, threshold-free cluster enhancement (TFCE) and 2D optimization for tract-
based DTI analysis. The CST region of interest (ROI) analysis was performed using the CST
mask (bilateral) included in the JHU white matter tractography atlas available in FSL,
thresholded at 0.5. The JHU-CST mask was further intersected with the study- specific
skeleton and the resulting mask was used for extracting the median values of DTI scalars in
the CST for each participant.
Cortical thickness and volumetric measures
For each patient cortical thickness of the bilateral precentral gyrus was obtained from the
native-space MPRAGE scans using the automated FreeSurfer 6.0 parcellation [18]. Total
brain volume (TBV), GM volume (GMV) and WM volume (WMV), normalized for head size,
were estimated using the SIENAX algorithm from the SIENA-package of FSL v5.0.
Results
Relationship between CSF NfL and DTI metrics across ALS phenotypes
Out of the whole sample n=29 classic ALS, n=14 LMND ALS and n=6 UMND ALS cases had
available both, measures of CSF NfL and DTI metrics. Unfortunately, in our cohort the group
of LMND ALS and UMND ALS was too small, lacking the power to perform phenotype-wise
analysis. However, correlation between DTI metrics in the CST and NfL level was present
also when restricting the analysis to the classic ALS cases (NfL and FA: rho=-0.4, p=0.03;
NfL and RD: rho=0.4, p=0.05; Supplemental Figure 2). Results in classic ALS are
convincing, overall supporting our main findings of a significant relationship between CSF
NfL and CST integrity.
Relationship between CSF NfL and survival across ALS phenotypes
The distribution of observed survival times over measured NfL levels is shown in
Supplemental Figure 3A for the distinct phenotypes. Classic phenotypes with survival times
greater than 8 years were excluded as they seem to exhibit a somewhat different course of
disease. Within that plot, the distribution of the values of the UMND ALS phenotype is
seemingly different from that of the scatter pattern of the classic phenotype. To elucidate this,
a principal component analysis with the variables NfL and survival time was performed
yielding the eigenvectors shown as arrows in black. The factor loads of [-0.7, 0.7] and [0.7,
0.7] lead to factor 1 describing constellations with low NfL values and comparatively longer
survival times and factor 2 pointing to individuals with longer survival times despite higher
NfL values. The scatter plot in factor coordinates in Supplemental Figure 3B reveals that
patients with the classic phenotype tend to scatter along the factor 1 axis, displaying low NfL
and relatively long survival times or for negative factor 1 values a combination of high NfL
with short survival times. The UMND group displays a negative mean for factor 1, so that
they also seem to exhibit elevated NfL levels corresponding to decreased survival times. But
this is somewhat offset by a positive mean value in their factor 2 components, allowing for
constellations with higher NfL levels than comparably long-lived classic cases or the ability to
survive longer than would be expected for a classic case with these NfL levels (or a
combination of those two). One may thus hypothesize that these results point to the