The usage of patient-derived skin fibroblasts in this study was approved by the Mayo Clinic Institutional Review Board. Skin fibroblasts obtained from CTRL individuals (no detectable CNS disorders) and patients diagnosed with ALS were obtained from the Mayo Clinic Center for Regenerative Medicine. Cells obtained from patients diagnosed with MS were obtained from the Mayo Clinic Center of Multiple Sclerosis and Autoimmune Neurology. Patient-derived skin fibroblasts were matched by age, sex, and passage number (Supplementary Table 1). All MS skin fibroblasts were collected from patients diagnosed with relapsing-remitting MS at the time of harvest. The time from diagnosis to skin fibroblast harvest ranged from 0.2 to 16.7 years in MS cells and 0.9 to 6.1 years in ALS cells (Supplementary Table 1). A total of 10 CTRL, 10 MS, and 10 ALS skin fibroblasts were used in this study (Supplementary Table 1). Any known family history of ALS is listed in Supplementary Table 1. Genotyping was not performed but does not exclude the possibility of possessing known genetic risk factors.

Minimum essential medium (MEM; cat. # 10-010-CV), phosphate-buffered saline (PBS; cat. # 21-031-CV), and MEM non-essential amino acids (NEAA; cat. # 25-025-CI) were purchased from Corning (New York, NY, USA). Trypsin (cat. # 25300), penicillin-streptomycin solution (PenStrep; cat. # 15140-122), and L-glutamine (L-Gln; cat. # 25030081) were purchased from Thermo Fisher Scientific (Waltham, MA, USA). Fetal bovine serum (FBS; cat. # F2442), silicone (cat. # Z273554), 10 mm × 10 mm cloning cylinders (cat. # CLS316610), and extracellular matrix gel (ECM gel; cat. # E1270) was purchased from Sigma-Aldrich (St. Louis, MO, USA). Low-e microscope slides were purchased from Kevley Technologies (cat. # CFR, Chesterland, OH, USA). Paraformaldehyde solution was purchased from Electron Microscopy Sciences (cat. # 15710, Hatfield, PA, USA).

Skin fibroblasts were incubated at 37°C in a humidified chamber with 5% CO₂/95% air maintained in MEM media containing 10% FBS, 2 mM L-Gln, 1X NEAA, and 1X PenStrep. Cells were passaged when approximately 80% confluent. At plating for experiments, all cell lines were within five passages of each other (passages 7–11, Supplementary Table 1).

Low-e slides were sterilized using 70% ethanol and air dried. Cloning cylinders were placed on the slide and sealed with silicone to prevent media from leaking. Each slide contained one CTRL, MS, and ALS skin fibroblast matched by age, sex, and passage number. Skin fibroblasts were detached from stock plates using trypsin, neutralized in media containing FBS, and spun. The supernatant was removed, and cells were suspended in media. A total of 20,000 cells in 200 μL of media were added per cloning cylinder. The slides were placed in a sterile 10 cm dish, covered, and incubated for 48 h. The media was removed, and skin fibroblasts were fixed in warm media containing 4% PFA for 10 min at 37°C. Cloning cylinders were removed and slides containing fixed cells were rinsed in PBS followed by Milli-Q water. Rinsed slides were dried in a glass desiccator containing desiccant for a minimum of 48 h.

All spectroscopic data was collected using an Agilent Cary 670 FTIR Spectrometer coupled to an Agilent Cary 620 FTIR Microscope (Agilent Technologies, Santa Clara, CA, USA) equipped with a liquid-N₂ cooled detector. The microscope stage is enclosed with a custom-built chamber to displace the atmosphere with a nitrogen stream. A 7 × 7 tile mosaic was collected with 128 × 128 focal plane array detector. The spectral resolution was set to 8 cm^–1 using a 25x objective with an optical resolution of 0.7 μm pixel size accumulating 80 scans per pixel. Measurements were done in reflection mode within the spectral region of 3500–950 cm^–1. Background measurements (120 scans per pixel) were taken from a clean area (no cells) on the slide using the same acquisition parameters and subtracted from the respective spectra to compensate for atmospheric interference and instrument performance. Scanning was carried out using Agilent Resolutions Pro Software (Agilent Technologies, Santa Clara, CA, USA).

The initial preprocessing of raw spectral files was carried out using Quasar based on Orange software (Demsar et al., 2013; Toplak et al., 2021), which utilized Python 3.8 software. Supplementary Figure 1 represents the general workflow used for pre-processing. In summary, wavenumbers 2700–2000 cm^–1 were removed due to the presence of a CO₂ peak in the absorbance spectra and low signal contribution of biological material (Bruun et al., 2006; Song et al., 2019). All remaining spectra were baseline corrected using the rubber band method. Next, we used unsupervised clustering (k-Means) to separate spectra of cells from near background measurements. The resulting spectra was binned to reduce the file size for further processing. To help correct for potential Mie scattering (Bassan et al., 2010; Baker et al., 2014; Troein et al., 2020), we utilized the software Open Chemometrics Toolbox for Analysis and Visualization of Vibrational Spectroscopy (OCTAVVS) (Troein et al., 2020), which utilized Python 3.8 software. The OCTAVVS platform uses an algorithm for clustered resonant Mie scattering correction, which greatly reduces computational time while achieving similar performance to previously described models (Troein et al., 2020). The absorbance spectra are corrected against a reference spectrum, which is typically produced from a homogenous sample (e.g., casein, Matrigel, etc.) (Bassan et al., 2010; Troein et al., 2020). For our study, we generated a reference spectrum using ECM gel for use in the OCTAVVS software (Supplementary Figure 2). The corrected spectra were then smoothed (Savitzky-Golay, windows 5, polynomial order 2), averaged, and vector normalized in Orange software. The averaged absorbance spectra were then used to generate second derivative spectra (Savitzky-Golay, windows 7, polynomial order 2, derivative 2) and vector normalized in Orange software.

The second derivative curves were used to help identify wavenumber regions and peak positions that correspond to biochemical features previously described in the literature (Talari et al., 2017). As pre-processing can affect the presence of spectral bands, we compared the second derivative results reported herein with those using a Savitzky-Golay windows 15 and polynomial order 5. Two biologically relevant peaks (_asN-CH₃ and _asCH₃, listed in Table 1) were apparent using the current analysis, which were reduced in the latter (data not shown). However, as a Savitzky-Golay of windows 7 and polynomial order 2 has been used to analyze skin, breast, and tonsil material (Piqueras et al., 2015; Valle et al., 2021; Shakya et al., 2022), and a polynomial order 2 has been recommended for FTIR spectra (Morais et al., 2019), we deemed the current analysis to be suitable. The wavenumber regions identified (Table 1) were used to integrate the respective area under the absorbance spectra (Baker et al., 2014; Cakmak-Arslan et al., 2020; Ferreira et al., 2020; Szentirmai et al., 2020). Significant changes in the integrated areas between groups were detected using a one-way ANOVA with post-hoc Tukey’s test (GraphPad Prism 9). Tests for normality (Shapiro–Wilk) and homoscedasticity (Brown–Forsythe) indicate that the data largely have a normal distribution and equal variances, respectively. Apart from the amide A variable of MS skin fibroblasts, which had a Shapiro-Wilk test P-value = 0.0279, all other P-values were not significant (P > 0.05). Sparse partial least squares-discriminant analysis (sPLS-DA) was performed using the R package mixOmics (Rohart et al., 2017). All averaged second derivative data (10 CTRL, 10 MS, and 10 ALS) were used for generating the sPLS-DA classification model using centroid-based distances. Data was divided into two sets for sPLS-DA using M-fold cross-validation (3-fold) repeated with 200 iterations. The optimum number of components was predicted to be three with 60, 40, and 20 latent variables on the first three components, respectively. We further tested the sPLS-DA model with a 2- and 3-fold repeated cross validation utilizing the second derivative data generated with a Savitzky-Golay window of 15 and polynomial order 5. The results were consistent with the former analysis and did not affect the conclusions as presented herein (data not shown). The sPLS-DA plot, AUROC curve plot, correlation plot, and loading plots were generated using the mixOmics package.

The absorbance spectra for all skin fibroblasts (10 CTRL, 10 MS, and 10 ALS, Supplementary Table 1) were collected in the 3500–950 cm^–1 range. All spectra were pre-processed using the same steps as outlined in the section “2. Materials and methods” and Supplementary Figure 1. The resulting absorption bands showed several characteristic peaks associated with biological samples (Figure 1A; Talari et al., 2017). To delineate specific wavenumber regions of interest for integrated analysis, we generated the second derivative spectra to help separate overlapping peaks (Figure 1B). Regions corresponding to amides, lipids, nucleic acids, and carbohydrates are highlighted in Figure 1 and listed in Table 1. Using the averaged second derivative spectra, multiple amide bands were detected including Amide A (3315–3260 cm^–1), Amide B (3120–3000 cm^–1), Amide I (1700–1580 cm^–1), Amide II (1580–1480 cm^–1), and Amide III (1300–1185 cm^–1). Regions mainly associated with lipids were detected at 3000–2800 cm^–1 (total lipids, symmetric and asymmetric CH₂/CH₃ vibrations), 1750–1725 cm^–1 [carbonyl esters, v(C = O)], and 1470–1430 cm^–1 [acyl chains, δ(CH₂)] (Snyder et al., 1996; Oleszko et al., 2015). The wavenumbers from 1300 to 1000 cm^–1 are composed of several metabolites and macromolecules including carbohydrates, proteins, lipids, and phosphate (e.g., nucleic acids, phospholipids, and phosphoproteins) (Colagar et al., 2011; Talari et al., 2017). These results demonstrate that patient-derived skin fibroblasts are suitable for generating spectral profiles as detected by FTIR spectroscopy.

In this study, we first aimed to determine if the spectral profiles from MS, ALS, and CTRL skin fibroblasts were unique as detected by FTIR spectroscopy. To do so, we analyzed the second derivative spectra (3500–2800 and 1800–950 cm^–1) using sparse partial least squares-discriminant analysis (sPLS-DA) to determine if the skin fibroblasts could be effectively classified into their respective groups (Figure 2). The sPLS-DA model is particularly useful for classification of FTIR spectroscopy data where high multi-collinearity exists amongst the variables and the number of variables far outnumber the samples while focusing on the discrimination between groups (Chung and Keles, 2010; Lê Cao et al., 2011; Rohart et al., 2017). Optimization of the model suggests that three components with 60, 40, and 20 variables, respectively, were optimal for group separation (Supplementary Figure 3). A 3D plot of the three components shows reasonable separation between MS, ALS, and CTRL skin fibroblasts (Figure 2A). Performance of the model was evaluated using an area under the receiver operating characteristics (AUROC) curve. The AUROC curve values were 1, 0.975 and 0.965 for CTRL, MS, and ALS, respectively, indicating good separation between groups (Figure 2B). Thus, these results suggest that second derivative spectra from patient-derived skin fibroblasts may be useful for the prediction and classification of diseased cells from MS, ALS, and CTRL individuals.

Quantitative ratios of the integrated absorbance spectra can provide insight into molecular changes occurring within tissue, cells, and biofluids, which have largely been described previously in the literature (Baker et al., 2014; Benseny-Cases et al., 2014; Talari et al., 2017). Additionally, ratios can help normalize and minimize the effect of experimental artifacts. Using the band regions identified from the second derivative spectra (Table 1), we calculated ratios of the respective integrated area under the absorbance spectrum (Figure 3 and Supplementary Table 2). We identified several lipid ratios that were significantly altered in MS and/or ALS skin fibroblasts compared to CTRL cells. The ratios of carbonyl ester/total lipids (∼1740/3000–2800 cm^–1), carbonyl ester/acyl chain (∼1740/1450 cm^–1), and carbonyl ester/asymmetric CH₃ (∼1740/2960 cm^–1) were significantly decreased in MS skin fibroblasts when compared to CTRL cells (Figures 3A–C). These altered ratios suggest that lipid homeostasis is perturbed, which may include abundance, membrane organization, structure, and peroxidation (Benseny-Cases et al., 2014; Oleszko et al., 2015; Barraza-Garza et al., 2016; Dreier et al., 2019). Interestingly, analysis of the integrated area under the absorbance spectra for total lipids, carbonyl ester, and acyl chains suggest an overall decrease in cellular lipids for MS and ALS skin fibroblasts when compared to CTRL cells (Supplementary Figures 4A–C). Additionally, ratios previously described to inform on membrane polarity (∼2920/2870 cm^–1), lipid chain packing (∼2850/2870 cm^–1), the degree of lipid saturation (∼2920/2960 cm^–1), and phospholipid chain length (∼2920/3000–2800 cm^–1) were further evaluated, however, we did not detect any significant changes (Supplementary Figures 4D–G; Saeed et al., 2015). These observations suggest that lipid abundance is reduced likely affecting cellular membrane organization, metabolism, signaling, and oxidative stress via peroxidation.

Similar to lipids, protein dysregulation in MS is well described (Andhavarapu et al., 2019). We analyzed several quantitative ratios related to protein dyshomeostasis. The ratio of Amide I to Amide II (1700–1580/1580–1480 cm^–1) revealed a significant increase in MS skin fibroblasts compared to CTRL cells (Figure 3D). A change in the Amide I/Amide II ratio is largely attributed to structural changes in proteins (Ricciardi et al., 2020). A general rise in protein abundance was detected in MS skin fibroblasts compared to ALS and CTRL cells as detected by an increase in the Amide I region (Supplementary Figure 4H) while Amide II levels were down in ALS (Supplementary Figure 4I). Interestingly, the ratio between Amide I and Amide A revealed a significant increase in MS compared to ALS (Figure 3E). Amide A largely arises from N-H stretching of peptide linkages, which is thought to arise from the overtone of the Amide II, and/or interactions between Amide I and Amide II, reflecting possible disorder in protein secondary structures (Krimm and Dwivedi, 1982; Magazù et al., 2012). Thus, the Amide I/Amide A ratio may be reflecting distinct protein structural changes in MS when compared to ALS skin fibroblasts. Overall, these findings suggest that protein structure, organization, and abundance are perturbed in MS and ALS skin fibroblasts compared to CTRL cells.

Lastly, we explored ratios linked to the status of membrane organization, protein folding, and cellular physiology (Szalontai et al., 2000; Colagar et al., 2011; Cakmak et al., 2012; El Khoury et al., 2019; Poonprasartporn and Chan, 2021). The Amide I/Total Lipids ratio resulted in a significant increase in MS skin fibroblasts compared to CTRL cells (Figure 3F). This reflects an imbalance in protein and lipid homeostasis. Similarly, the Amide I/carbonyl ester is significantly increased in MS skin fibroblasts (Figure 3G), which may affect membrane organization and protein folding and dynamics. The region 1300–1000 cm^–1 is comprised of lipids, proteins, phosphate, and carbohydrates reflecting several metabolites and macromolecules (Colagar et al., 2011; Caine et al., 2012). The ratio of carbonyl ester/1300−1000 cm^–1 was found to be significantly decreased in both MS and ALS skin fibroblasts compared to CTRL cells suggesting diseased cells are biochemically distinct (Figure 3H). Taken together, these observations indicate that the physiological status of diseased skin fibroblasts are unique suggesting that cellular membranes, protein dynamics, oxidative stress, and metabolite profiles are perturbed.

The second derivative spectra are particularly useful to interrogate shifts in the peak location and shape, which can help inform on biomolecular alterations within the biospecimens. Therefore, we compared the mean second derivative spectra of MS, ALS, and CTRL skin fibroblasts (Figure 4). Notably, we detected a shift in the peak positions of the Amide I and Amide II bands (Figures 4A, B, respectively). The Amide I band near the α-helical region (∼1650 cm^–1) in MS and ALS shifted toward higher wavenumbers when compared to CTRL skin fibroblasts (Figure 4A). This suggests an increase in β-turn structures in both MS and ALS cells, which may be indicative of protein alterations commonly associated with neurological disorders including MS and ALS (David and Tayebi, 2014; McAlary et al., 2019; Ricciardi et al., 2020). In the Amide II region, we detected three peaks (Figure 4B). The peaks near 1545 and 1520 cm^–1 are largely associated with α-helixes and β-sheets, respectively (Murphy et al., 2014). The third peak around 1490 cm^–1 is less clear but may be related to additional β-structures or choline groups of lipids (Casal and Mantsch, 1984; Murphy et al., 2014; Oleszko et al., 2015). In agreement with the shift in the α-helical region of the Amide I peak, the α-helix peak detected in the Amide II region was also shifted toward higher wavenumbers in MS and ALS when compared to CTRL skin fibroblasts further suggesting altered secondary structures in the diseased cells (Figure 4B). The Amide A and Amide B peaks were found between ∼3350 and 3000 cm^–1 (Figure 4C). While the Amide B peak was relatively weak, we observed a prominent band for Amide A. An apparent increase in the Amide A intensity and shift toward lower wavenumbers was observed in both MS and ALS skin fibroblasts when compared to CTRL cells (Figure 4C), which is likely associated with the α-helix shifts observed in the Amide I and II regions (Magazù et al., 2012). Four peaks were observed in the lipid region (3000–2800 cm^–1) mainly arising from C-H stretching including the symmetric CH₂ [v_s(CH₂), 2860–2843 cm^–1], symmetric CH₃ [v_s(CH₃), 2880–2865 cm^–1], asymmetric CH₂ [v_as(CH₂), 2936–2912 cm^–1], and asymmetric CH₃ [v_as(CH₃), 2972–2950 cm^–1] peaks (Figure 4D). Two additional peaks, carbonyl ester and acyl chain, were detected in skin fibroblasts, which are thought to arise predominately from lipids (Derenne et al., 2013). The carbonyl ester [v(C = O)] was detected around 1735 cm^–1 and the acyl chain near 1450 cm^–1 (Figures 4E, F, respectively). The acyl chain of CTRL cells reveals two peaks likely representing the bending/scissoring of CH₂ [∼1455 cm^–1, δ(CH₂)] and deformation mode of methyl groups [∼1435 cm^–1, δ_as(CH₃)] (Casal and Mantsch, 1984; Ami et al., 2014). Interestingly, MS and ALS skin fibroblasts have a single acyl chain peak possibly indicating altered organization, structure, and packing of the acyl chains in lipid membranes (Figure 4F; Casal and Mantsch, 1984; Ami et al., 2014). Overall, the second derivative spectra suggests that protein structure and lipid organization are altered in MS and ALS cells when compared to CTRL skin fibroblasts.