*Article* **In Vitro and In Vivo Effects of SerpinA1 on the Modulation of Transthyretin Proteolysis**

**Filipa Bezerra 1,2, Christoph Niemietz 3, Hartmut H. J. Schmidt 3,†, Andree Zibert 3, Shuling Guo 4, Brett P. Monia 4, Paula Gonçalves 1, Maria João Saraiva 1,2 and Maria Rosário Almeida 1,2,\***


**Abstract:** Transthyretin (TTR) proteolysis has been recognized as a complementary mechanism contributing to transthyretin-related amyloidosis (ATTR amyloidosis). Accordingly, amyloid deposits can be composed mainly of full-length TTR or contain a mixture of both cleaved and full-length TTR, particularly in the heart. The fragmentation pattern at Lys48 suggests the involvement of a serine protease, such as plasmin. The most common TTR variant, TTR V30M, is susceptible to plasmin-mediated proteolysis, and the presence of TTR fragments facilitates TTR amyloidogenesis. Recent studies revealed that the serine protease inhibitor, SerpinA1, was differentially expressed in hepatocyte-like cells (HLCs) from ATTR patients. In this work, we evaluated the effects of SerpinA1 on in vitro and in vivo modulation of TTR V30M proteolysis, aggregation, and deposition. We found that plasmin-mediated TTR proteolysis and aggregation are partially inhibited by SerpinA1. Furthermore, in vivo downregulation of SerpinA1 increased TTR levels in mice plasma and deposition in the cardiac tissue of older animals. The presence of TTR fragments was observed in the heart of young and old mice but not in other tissues following SerpinA1 knockdown. Increased proteolytic activity, particularly plasmin activity, was detected in mice plasmas. Overall, our results indicate that SerpinA1 modulates TTR proteolysis and aggregation in vitro and in vivo.

**Keywords:** transthyretin; SerpinA1; ATTR amyloidosis; TTR proteolysis; plasmin

### **1. Introduction**

Transthyretin-related amyloidoses (ATTR amyloidosis) are characterized by extracellular deposition of insoluble TTR amyloid fibrils in several tissues, being polyneuropathy and cardiomyopathy the major clinical manifestations, as reviewed in [1,2]. There are two types of ATTR amyloidoses: hereditary amyloidosis (ATTRm) and wild-type ATTR amyloidosis (ATTRwt) [3]. ATTRm occurs through single-residue substitutions in TTR, mainly producing less stable variants [4–6], whereas ATTRwt is an age-related disorder, affecting 20–25% of the population over 80 years, with predominant cardiac phenotype, characterized by wild-type (WT) TTR amyloid deposits [7,8].

Despite it is widely accepted that tetramer destabilization is a rate-limiting step for the development of amyloid fibrils [9–12], TTR proteolysis has been increasingly recognized as another mechanism driving TTR amyloid formation. Several studies reported the existence of different TTR amyloid deposits in a range of tissues. Thus, amyloid deposits might be composed mainly of full-length TTR (type-B fibrils) or by a mixture of both

**Citation:** Bezerra, F.; Niemietz, C.; Schmidt, H.H.J.; Zibert, A.; Guo, S.; Monia, B.P.; Gonçalves, P.; Saraiva, M.J.; Almeida, M.R. In Vitro and In Vivo Effects of SerpinA1 on the Modulation of Transthyretin Proteolysis. *Int. J. Mol. Sci.* **2021**, *22*, 9488. https://doi.org/10.3390/ ijms22179488

Academic Editor: Masaru Tanaka

Received: 20 July 2021 Accepted: 28 August 2021 Published: 31 August 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

cleaved and full-length TTR (type-A fibrils) [13–16]. Type-A fibrils occur in several tissues, particularly in the heart and are related to the development of restrictive cardiomyopathy after liver transplantation, leading to poor clinical outcomes in ATTR V30M patients [17]. The protease responsible for TTR cleavage has not yet been identified. However, the specific fragmentation at Lys48 in the TTR polypeptide suggests that it could be a trypsin-like serine protease [18]. In vitro experiments using recombinant trypsin indicate that several amyloidogenic TTR variants are susceptible to trypsin-mediated proteolysis [19–21]. The process of cleavage and release of the 49–127 TTR fragment, the most frequent fragment detected in fibrils, is faster for the highly amyloidogenic variant, TTR S52P, being the 49–127 C-terminal fragment rapidly incorporated into amyloid fibrils [19,20].

Additionally, in silico studies pointed out plasmin as a plausible pathophysiological candidate protease involved in the process of TTR amyloid formation [22]. Furthermore, the ubiquitous distribution of plasmin, its structural similarities to trypsin [22], and the reported activation of plasminogen activation system (PAS) in other amyloid-related disorders, such as Alzheimer's disease [23] and immunoglobulin light chain (AL) amyloidosis [24–26] indicate that this protease could have a key role in TTR amyloidogenesis. Indeed, the 49–127 C-terminal TTR fragment was also found in in vitro plasmin-digested TTR S52P samples, suggesting that TTR-digested samples are more prone to aggregate than the non-digested ones [22].

On the other hand, serine protease inhibitors (Serpins), particularly SerpinA1, have also been related to pathological proteinopathies, such as Alzheimer's disease [27–29]. We hypothesized that SerpinA1 may act as a modulator of TTR proteolysis/fibrillogenesis. SerpinA1 belongs to the clade A serpins, which are classified as antitrypsin-like [30]. SerpinA1 and also SerpinA3 were found to be differentially expressed in ATTRm patients compared to healthy controls [31,32]. The SerpinA1 mRNA was found to be differentially expressed in hepatocyte-like cells (HLCs) from ATTR patients compared to healthy controls and, a high inverse correlation between *SerpinA1* and *TTR* genes was also observed. Upon TTR knockdown in HLCs the correlation was abolished [33]. Recently, it was demonstrated that SerpinA1 knockdown modulates TTR expression in both cellular and animal models, performing an important role in the development of ATTR amyloidosis [34].

In this study, we further explored the role of SerpinA1 on the in vitro and in vivo modulation of plasmin-mediated TTR proteolysis and how this modulation may impact TTR amyloidogenesis to contribute to the development of more targeted therapies for the treatment of ATTR amyloidosis.

#### **2. Results**

#### *2.1. SerpinA1 Inhibits In Vitro Plasmin-Mediated Proteolysis of Transthyretin V30M*

Previous studies revealed that similarly to TTR S52P, recombinant TTR V30M was also susceptible to plasmin-mediated proteolysis [22]. In addition, amyloid deposits extracted from cardiac and adipose tissue specimens from ATTR V30M patients were composed of a mixture of both cleaved and full-length TTR [13–15]. Thus, we performed in vitro proteolysis experiments to evaluate whether SerpinA1 performs a role on the inhibition of plasmin-mediated TTR V30M proteolysis. Recombinant TTR V30M was incubated with plasmin at 37 ◦C for 24 h, and the resulting mixture was analyzed by SDS-PAGE and Western blot. Besides the dimeric (~35 kDa) and monomeric (~17 kDa) TTR V30M forms, Western blotting analysis revealed the presence of two different TTR fragments, in contrast to non-digested samples (Figure 1A and Supplementary data Figure S1). These fragments were detected with the commercially available antibody produced and characterized in our lab [35], anti-TTR mutant (Y78F), clone AD7 (Figure 1A) but not with rabbit polyclonal anti-TTR from DAKO (Figure 1B).

**Figure 1.** Plasmin cleaves transthyretin V30M, and its activity is inhibited by SerpinA1. Representative images of Western blotting analysis of the three independent in vitro experiments. Western blot was performed using two different antibodies targeting human TTR, mouse anti-TTR mutant (Y78F), clone AD7 (**A**), and rabbit anti-transthyretin (**B**). Both antibodies detected dimeric and monomeric TTR forms. However, only the mouse anti-TTR mutant (Y78F) clone AD7 detects TTR fragments (**A**).

N-terminal sequencing analysis of the TTR fragments firstly observed in Western blotting analysis indicates that band 1 corresponds to a peptide starting at position 49 and band 2 to a peptide starting at the first amino acid of TTR polypeptide chain (Table 1). Furthermore, the bands corresponding to TTR fragments were excised from SDS-PAGE gels and further analyzed by mass spectrometry (MS) analysis after trypsin digestion. These MS experiments revealed that band 1 was composed of the peptides with mass corresponding to the amino acids 81–103, 104–127, 105–126, and 105–127 (Table S1), which along with N-terminal sequencing data, indicated that band 1 should correspond to the TTR fragment 49–127 (Table 1). Band 2 was composed by the peptides with mass corresponding to the amino acids 1–15, 22–34, 35–48, and 36–48 (Table S1). Together with N-terminal sequencing data, our results indicate that band 2 was the TTR fragment 1–48 (Table 1). Intriguingly, the band containing 1–48 N-terminal TTR fragment also revealed the presence of a C-terminal peptide comprising the amino acids 105–127. Since the number of peptidespectrum matches (PSMs) identified for that peptide group was only two, which was too low compared to the N-terminal peptides, this might indicate that this C-terminal peptide 105–127 was a contaminant.

**Table 1.** Identification of transthyretin peptides upon plasmin digestion by N-terminal sequencing. Band 1 starts at the amino acid residue 49, whereas band 2 starts at the first amino acid residue in the transthyretin polypeptide chain, indicating that band 1 corresponds to the 49–127 C-terminal fragment and band 2 to the 1–48 N-terminal fragment.


Following, we investigated the role of the serine protease inhibitor, SerpinA1, as a modulator of plasmin-mediated TTR proteolysis. We found that TTR proteolysis was partially inhibited in the presence of SerpinA1 (Figure 1A and Figure S1). No TTR fragments were observed neither before the assay nor in the absence of plasmin, excluding the influence of TTR auto-proteolysis or degradation. In opposition to TTR V30M, TTR WT was not susceptible to plasmin-mediated proteolysis under the same conditions as presented in supplementary data (Figure S2).

### *2.2. SerpinA1 Inhibits In Vitro Transthyretin V30M Aggregation upon Plasmin-Mediated Proteolysis*

The presence of TTR fragments, particularly the 49–127 C-terminal peptide, was implicated in TTR amyloidogenesis. Studies of in vitro TTR cleavage indicated that this fragment generated upon digestion with trypsin or plasmin was rapidly incorporated into amyloid fibrils, suggesting that TTR proteolysis facilitated the process of TTR aggregation [19–22]. Similarly, in the present work, we investigated the influence of plasminmediated proteolysis on the aggregation potential of TTR V30M. Upon 24 h of incubation, both plasmin-digested and non-digested samples were characterized using dynamic light scattering (DLS) analysis (Figure 2). Plasmin-mediated proteolysis facilitates the process of TTR aggregation, as observed by the increase of TTR aggregated species (909.7 nm; 8.1%) along with the decrease of the soluble form (13.15 nm; 75.4%) (Figure 2B), comparatively to non-digested samples, in which TTR was only found in the soluble form (9.228 nm; 98.5%) (Figure 2A). In addition, TTR V30M incubated with SerpinA1 revealed the presence of soluble particles exhibiting a large diameter (23.81 nm, 100%) (Figure 2C), probably indicating the formation of TTR-SerpinA1 complex, as described previously [33]. Samples incubated with both plasmin and SerpinA1 presented less abundant and smaller TTR aggregates (537.7 nm; 4.3%) (Figure 2D), as compared to samples only incubated with plasmin (Figure 2B).

**Figure 2.** Plasmin enhances the transthyretin V30M aggregation, while SerpinA1 inhibits the process. Dynamic light scattering analysis was performed upon transthyretin V30M incubation at 37 ◦C for 24 h: (**A**) alone; (**B**) with plasmin; (**C**) with SerpinA1; (**D**) with plasmin and SerpinA1. The data results from three independent in vitro experiments, each one performed in triplicate.

Besides its function as a serine protease inhibitor, SerpinA1 also functions as an extracellular chaperone and recently, it was reported that SerpinA1 inhibited TTR amyloid formation in vitro [33]. Thus, the same samples were analyzed by thioflavin T (ThT) assays to evaluate the amyloid nature of the formed species. The results demonstrated that plasmin facilitates TTR V30M amyloid formation, as observed by the increased ThT emission fluorescence signals upon plasmin incubation (1086 ± 75.68 vs. 901.3 ± 91.88 in the absence of plasmin) (Figure 3). Moreover, in the presence of plasmin, TTR V30M amyloid formation was significantly inhibited by SerpinA1 (510 ± 60.58 vs. 1086 ± 75.68 in the absence of SerpinA1; *p* < 0.05). SerpinA1 per se seemed to inhibit TTR V30M amyloid formation (613 ± 134 vs. 901.3 ± 91.88 in the absence of SerpinA1) (Figure 3).

**Figure 3.** Transthyretin amyloid formation is favored upon plasmin-mediated proteolysis, being partially inhibited by SerpinA1. Thioflavin T experiments were performed upon transthyretin V30M incubation with plasmin and/or SerpinA1, at 37 ◦C for 24 h. The fluorescence emission signal of thioflavin T of transthyretin V30M at the beginning of the experiment (t = 0 h) is represented as the red dotted line around 607. The data results from three independent in vitro experiments (*n* = 3). Statistical analysis was performed using one-way ANOVA with Tukey's multiple comparison as post-test. \* *p* < 0.05 in the presence of plasmin/presence of SerpinA1 vs. presence of plasmin/absence of SerpinA1, q = 6.092, df = 3. Effect size (r) = 0.924, odds ratio = 0.69 for an interval of confidence of 95%.

### *2.3. SerpinA1 Downregulation Increased Transthyretin Deposition in the Heart of Old Transgenic Mice Carrying Human Transthyretin V30M Mutation*

Previous studies reported that SerpinA1 knockdown was accompanied by an increase in TTR mRNA expression, as well as TTR protein levels in HepG2 cells. In collaboration with our group, it was also reported that SerpinA1 knockdown resulted in an increase in TTR mRNA expression in mouse liver, as well as in TTR protein levels in plasmas of transgenic mice carrying human TTR V30M mutation (HM30) [34]. Therefore, we decided to investigate whether SerpinA1 was also specifically downregulated in the mouse heart and whether effects on TTR protein levels could be observed. For that, SerpinA1-specific ASOs were subcutaneously administered to HM30 mice once a week for six weeks. Western blotting analysis (Figure 4A–C), as well as immunohistochemistry (Figure 4D), confirmed that SerpinA1 was effectively downregulated in the heart from younger and older animals (0.018 ± 0.01 vs. 1.0 ± 0.18 in ASO-CTR, *p* < 0.0001). In addition, a significant increase in TTR protein in mice cardiac tissue was observed (1.6 ± 0.18 vs. 1.0 ± 0.14 in ASO-CTR group, *p* = 0.020) (Figure 4E–G).

**Figure 4.** Human transthyretin was found to be increased in the heart of transgenic mice carrying human transthyretin V30M mutation upon SerpinA1 downregulation. Western blotting analysis of SerpinA1 and GAPDH expression in the heart of old (16–21 months, *n* = 7) (**A**) and young (12–13 months, *n* = 12) (**B**) mice. Quantification of SerpinA1 expression normalized to GAPDH (loading control protein) of the two pooled experiments (**C**). Immunohistochemistry data also reveal that SerpinA1 was effectively downregulated in the heart of HM30 mice. Images were captured at 10× magnification using Olympus BX50 microscope. Scale bar = 20 μm (**D**). Western blotting of TTR and GAPDH expression in mice cardiac tissue of both old (**E**) and young (**F**) animals. Bar plot represents the quantification of transthyretin expression normalized to GAPDH of the two pooled in vivo experiments (**G**). Protein bands were quantified by densitometry using Image J. Statistical analysis was performed using an unpaired *t*-test. \*\*\* *p* < 0.0001 when comparing the SerpinA1/GAPDH ratio between ASO-CTR and mA1AT-ASO, t = 6.026, df = 17; \* *p* < 0.05 when comparing the hTTR/GAPDH ratio between ASO-CTR and mA1AT-ASO, t = 2.554, df = 17. Calculations were performed for an interval of confidence of 95%.

Our previous data revealed that, along with increased TTR mRNA and protein levels, SerpinA1 knockdown increased in vivo TTR deposition in several tissues, such as dorsal root ganglia (DRGs) and intestine [34]. In this study, we addressed whether TTR deposition was also promoted in mouse cardiac tissue after SerpinA1 downregulation. In younger animals, duodenal TTR deposition was found to be increased upon silencing of SerpinA1 (1.773 ± 0.94 vs. 0.1951 ± 0.04 in ASO-CTR, *p* = 0.0087) (Figure 5, upper panel), and a similar tendency was also observed in older animals (Figure 5, lower panel). Additionally, immunohistochemistry analysis demonstrated that TTR deposition was favored in the heart of older animals upon SerpinA1 downregulation (8.307 ± 0.6697 vs. 3.049 ± 1.269 in ASO-CTR, *p* = 0.0106) (Figure 5, lower panel), comparatively to the younger group (Figure 5, upper panel).

**Figure 5.** Transthyretin deposition is increased upon SerpinA1 knockdown in the mouse heart of older animals. Representative images of immunohistochemical analysis of mouse heart and duodenum upon the administration of antisense oligonucleotides targeting SerpinA1 to young

(12–13 months, *n* = 12) (upper panel) and old (16–21 months, *n* = 7) (lower panel) animals. Bar plot representation of TTR quantification normalized to the total occupied area in the duodenum (**A**,**C**) and heart (**B**,**D**) of young (**A**,**B**) and old (**C**,**D**) mice. Images were captured at 10× magnification using Olympus BX50 microscope and analyzed using Image Pro Plus software. Scale bar = 20 μm. Statistical analysis was performed using an unpaired *t*-test. \*\* *p* < 0.01 comparing TTR deposition in the duodenum of younger animals in ASO-CTR vs. mA1AT-ASO, t = 3.300, df = 10; \* *p* < 0.05 comparing TTR deposition in the heart of younger animals in ASO-CTR vs. mA1AT-ASO, t = 3.969, df = 5. Calculations were performed for an interval of confidence of 95%.

### *2.4. Transthyretin Fragments Are Observed in Mouse Cardiac Tissue upon SerpinA1 Downregulation*

Based on our experiments of in vitro plasmin-mediated proteolysis described above, indicating that SerpinA1 inhibits TTR cleavage, we evaluated the impact of SerpinA1 knockdown on TTR cleavage in vivo. Western blotting analysis revealed the presence of a protein band below to the monomeric TTR, corresponding to TTR fragment (<11 kDa) in mouse cardiac tissue in younger (Figure 6A) and older animals (Figure 6B). Similarly, these fragments were only detected using the antibody anti-TTR mutant (Y78F), clone AD7. In opposition, no TTR fragments were found upon SerpinA1 knockdown neither in other tissues of TTR deposition, such as duodenum (Supplementary data Figure S4A) and stomach (Supplementary data Figure S4B) nor in mice plasmas (Supplementary data Figure S4C).

### *2.5. SerpinA1 Downregulation Increases Proteolytic Activity, Particularly Plasmin Activity, in Plasmas of Transgenic Mice Carrying Human Transthyretin V30M Mutation*

SerpinA1 partially inhibits plasmin-mediated proteolysis in vitro. Thus, fluorescencebased enzymatic assays were performed to evaluate the effects of downregulation of SerpinA1 on serine protease activity, namely plasmin activity in vivo. In fact, serine protease activity was effectively increased in plasmas of HM30 mice upon SerpinA1 downregulation (12.69 ± 1.746 vs. 9.606 ± 1.675 in ASO-CTR, *p* = 0.046) (Figure 7A) while no proteolytic activity was found in mice heart homogenates (Figure S5). In particular, the activity of plasmin was also found to be increased in mice plasmas after SerpinA1 knockdown (8422 ± 432 vs. 7013 ± 458 in ASO-CTR, *p* = 0.040) (Figure 7B).

**Figure 6.** Western blotting analysis revealed the presence of transthyretin fragments upon SerpinA1 downregulation in the heart. Transthyretin fragments were detected in the heart of both older (**A**) and younger (**B**) animals.

**Figure 7.** Serine protease activity and, particularly plasmin activity was found to be increased in mice plasmas upon SerpinA1 knockdown. Serine protease activity (**A**) and plasmin activity (**B**) were measured in plasma samples of HM30 mice according to the manufacturer's instructions. Statistical analysis was performed using an unpaired *t*-test. \* *p* < 0.05 comparing protease activity between ASO-CTR and mA1AT-ASO, t = 2.189, df = 14; \* *p* < 0.05 comparing plasmin activity between ASO-CTR and mA1AT-ASO, t = 2.235, df = 16. Calculations were performed for an interval of confidence of 95%.

#### **3. Discussion**

Tetramer destabilization is considered the rate-limiting step driving TTR amyloidogenesis. However, TTR proteolysis has been reported as an additional mechanism contributing to TTR amyloid formation [36]. The 49–127 C-terminal fragment is the most frequently encountered in ex vivo amyloid fibrils [13,15], and, the fragmentation pattern at Lys residues indicates the activity of a trypsin-like serine protease [18,37].

Bellotti and collaborators identified three TTR fragments upon in vitro incubation of the highly amyloidogenic TTR S52P with trypsin, being the 49–127 C-terminal fragment more prone to aggregation than the 16–127 and 81–127 C-terminal fragments [19]. A similar pattern of fragmentation was obtained with plasmin, a ubiquitous, widely distributed serine protease related to fibrinolysis [22].

Recently, the serine protease inhibitor, SerpinA1, was implicated in ATTR amyloidogenesis, and we hypothesized that it could also modulate TTR proteolysis. Previous studies by Niemietz et al. demonstrated that SerpinA1 inhibited TTR aggregation both in vitro, in cell culture experiments using hepatocyte-like cells (HLCs), and in vivo, in a study of SerpinA1 knockdown in mice carrying human TTR V30M mutation (HM30) [33,34].

In this study, our aim was to investigate the role of SerpinA1 as an inhibitor of serine proteases and its effect on the in vitro and in vivo modulation of TTR amyloid formation to contribute to a better knowledge of the process and to search for new and more specific therapeutic approaches.

In this sense, we confirmed that TTR V30M, the most frequent TTR variant related to ATTR amyloidosis, was prone to plasmin-mediated proteolysis in vitro and, that the cleaved protein aggregates more rapidly than the non-cleaved TTR V30M. N-terminal sequencing and MS analysis of the bands corresponding to TTR fragments generated by plasmin-mediated proteolysis identified the presence of the peptides 1–48 N-terminal and 49–127 C-terminal. The N-terminal region of the TTR polypeptide chain was enriched in hydrophobic amino acid residues and, importantly, the 26–57 TTR segment, belonging to the aggregation-prone regions (APR), exhibited high amyloid propensity [18]. These APR were protected when the protein was in its native form [38]. However, the destabilization of the native TTR structure induced by the single-point mutation at position 30 might expose those regions to cleavage, and, for that, the fragment 1–48 N-terminal TTR fragment may also be potentially considered highly amyloidogenic. It has been demonstrated that the 49–127 C-terminal fragment facilitates TTR amyloid formation in vitro [19–22], and, accordingly, Dasari et al. recently determined that the proteolytic cleavage of the K48-T49 peptide bond in the CD loop accelerated the formation of small spherical oligomers, which exhibited cytotoxic effects in neuroblastoma SH-SY5Y cells [39]. It was also shown that TTR aggregates generated by full-length or truncated TTR forms exhibited nearly identical molecular structural features, suggesting that TTR proteolysis in the CD loop destabilizes the native TTR tetramer. This destabilization of the TTR tetramer promotes oligomer formation through a similar mechanism of TTR misfolding and aggregation rather than through another molecular mechanism [39].

Marcoux et al. suggested the influence of biomechanical forces, particularly shear stress forces generated by fluid flow, on TTR proteolysis, which could influence the tissue specificity of TTR amyloid deposition. Indeed, a mechano-enzymatic cleavage mechanism for TTR proteolysis was proposed, where tetrameric TTR might be cleaved prior to TTR deposition and, then, due to strong shear stress observed in the heart, the C-terminal fragments would be released being rapidly incorporated into amyloid fibrils. Alternatively, both cleavage and dissociation may occur simultaneously at the heart, where both local shear stress forces and the relevant protease could be present [20]. These shear and interfacial forces are particularly strong in the cardiac tissue [40], which might explain the frequently encountered type-A fibrils in TTR deposits found in the heart. However, the presence of type-A amyloid fibrils in other tissues, such as the vitreous humor and the spinal cord of ATTR V30M patients, indicate that this mechanism would not explain the formation of TTR amyloid deposits based on their tissue-specific location, since these shear stress conditions were not observed neither in the eye nor in the central nervous system [41–43]. Moreover, in a recent study of Suhr et al., 14 out of 15 families with ATTR V30M amyloidosis exhibited a similar amyloid fibril composition within family members, independently of the age-onset disease. These observations indicate that, besides specific tissue/organ characteristics, genetic and/or epigenetic alterations may influence the amyloid fibril composition [44].

Our in vitro results using recombinant TTR show that SerpinA1 partially inhibits plasmin-mediated TTR proteolysis and suggest that, in parallel, it can also have an effect on the inhibition of TTR V30M amyloid formation seem to be independent of the presence of plasmin. Interestingly, this effect was compatible with the physical interaction between SerpinA1 and TTR that was recently suggested [33]. In addition, our DLS data indicate the presence of large diameter soluble particles, possibly the SerpinA1-TTR complex. Thus, future studies should be performed to clarify whether SerpinA1 performs an important role as a modulator of TTR proteolysis through its interaction with TTR, avoiding the access of plasmin to its targeting region in the TTR structure.

Our previous studies of SerpinA1 downregulation showed significantly increased TTR serum levels in HM30 mice, as well as in hepatoma cells [34]. Furthermore, SerpinA1 knockdown led to increased TTR deposition in the gastrointestinal tract, as well as in the sciatic nerve and dorsal root ganglia (DRG) of HM30 mice. In this work, we also found increased TTR protein deposits in the heart of older HM30 mice, whereas increased duodenal TTR deposition was found in the younger mice and, the same tendency was also observed in older ages. Moreover, we detected TTR fragments in mouse cardiac tissue upon SerpinA1 downregulation, while no TTR fragments were detected in mice plasmas nor in deposits from other tissues, such as the duodenum and stomach. The absence of fragments in mice plasmas upon SerpinA1 knockdown, as revealed in AD7 immunoblot, might be related to a very low concentration of TTR fragments in plasma and/or to insufficient sensitivity of the method. Additionally, we found increased serine protease activity, particularly plasmin activity, in plasmas upon treatment with ASOs targeting SerpinA1, whereas no proteolytic activity was observed in the heart of HM30 mice (Figure S3).

Despite the increasing interest in TTR proteolysis as a leading mechanism-driving TTR amyloidosis, some questions remain to be answered, namely whether TTR fragmentation occurs, prior to or after TTR aggregation/deposition and where it occurs. Some authors reported that plasmin degrades amorphous protein aggregates, releasing smaller soluble protein fragments, which were cytotoxic to both endothelial and microglial cells [45]. Trypsin or trypsin-like enzymes directly cleave acid-induced aggregates of full-length TTR V30M and barely cleave native soluble TTR V30M tetramer [46]. Additionally, recent cryo-electron microscopy (cryo-EM) experiments revealed the co-existence of both N-terminal and C-terminal TTR segments in one TTR fibril and, the relative special arrangement of these two segments are compatible with full-length TTR, suggesting that the process of fibril formation precedes TTR proteolysis [47]. In opposition, other studies showed increased proteolytic activity in plasmas from ATTR patients compared to healthy controls, suggesting that TTR proteolysis occurs in the bloodstream prior to TTR aggregation/deposition [31]. Accordingly, our data demonstrating increased protease and plasmin activity in mice plasmas, along with the absence of protease activity in mice hearts, suggest that in vivo TTR proteolysis occurs before fibril formation (Figure 7A,B, and supplementary data Figure S5).

In summary, our in vitro experiments demonstrate that plasmin cleaves the recombinant TTR V30M proteolytically and promotes its aggregation in vitro. Additionally, SerpinA1 partially inhibits the activity of plasmin in vitro, which decreases TTR amyloid formation. To investigate the relevance of these findings in vivo, SerpinA1 expression was knockdown in HM30 mice. The absence of SerpinA1 favored TTR deposition in mice tissues and increased the serine protease activity, namely plasmin activity in mice plasmas, which was accompanied by the presence of TTR fragments in the mice heart.

This work presents some limitations in particular concerning the knowledge of the detailed mechanism involving SerpinA1 inhibition of plasmin-mediated TTR proteolysis and also the impact of SerpinA1 downregulation in mice carrying TTR WT or carrying non-V30M TTR mutations. Accordingly, future experiments must be performed namely to dissect the molecular mechanisms by which SerpinA1 inhibits plasmin activity through direct interaction with the protease or by the formation of TTR-SerpinA1 complex. Additionally, it would also be interesting to evaluate the role of plasmin on TTR proteolysis using different animal models developing ATTR amyloidosis, such as transgenic mice carrying human A97S mutation [48]. Ultimately, it is important, to investigate plasmin and other extracellular serine proteases activity in TTR V30M patients and patients carrying other TTR amyloidogenic mutations to evaluate whether this activity has tissue-specific effects or is related to disease progression potentiating its interest as a biomarker in ATTR amyloidosis.

Altogether our in vitro and in vivo results show that plasmin is a plausible protease performing a role on TTR proteolysis and reveal SerpinA1 as an important modulator of the process of TTR cleavage. Our findings might contribute to the development of more effective and targeted therapies for the treatment of ATTR amyloidosis.

### **4. Materials and Methods**

#### *4.1. Reagents*

Native human plasmin protein (active), native human SerpinA1 protein (active), protease activity assay kit, plasmin activity assay kit, and recombinant rabbit anti-GAPDH antibody were purchased from Abcam (Cambridge, UK). Mouse monoclonal antibody anti-TTR mutant (Y78F), clone AD7 was from Merck Millipore (Sigma-Merck, Darmstadt, Germany). Rabbit polyclonal anti-human TTR antibody was from DAKO (Hovedstaden, Denmark). Rat monoclonal anti-mouse SerpinA1 antibody was from R&D systems (Minneapolis, MN, USA). Pierce TM High Capacity Endotoxin Removal Resin was from Thermo Scientific (Waltham, MA, USA). GalNAc-AAT ASO (mA1AT-ASO: ACCCAATTCAGAAG-GAAGGA) and GalNAc-Control ASO (ASO-CTR: CCTTCCCTGAAGGTTCCTCC) [48] were kindly provided by Ionis Pharmaceuticals (Carlsbad, CA, USA).

### *4.2. Recombinant Human Transthyretin*

TTR V30M and TTR WT were produced using a bacterial expression system and purified as previously described [49]. Recombinant TTR was applied to an affinity chromatography column packed with Pierce TM High Capacity Endotoxin Removal to remove bacterial lipopolysaccharides according to the manufacturer's instructions. TTR was then dialyzed against endotoxin-free phosphate-buffered saline (PBS) (Sigma-Merck, Darmstadt, Germany), concentrated using Vivaspin ultrafiltration units (GE Healthcare, Chicago, IL, USA), and quantified using Bradford protein assay (Bio-Rad, Hercules, CA, USA).

#### *4.3. In Vitro Plasmin-Mediated Proteolysis Assays*

TTR variants were firstly filtered through a sterile 0.2 μm inorganic membrane AN-OTOP syringe filter (Whatman, Maidstone, UK) to remove any protein aggregates. Then, TTR (18 μM) was incubated with plasmin (0.4 U) and/or SerpinA1 (2.8 μM) at 37 ◦C for 24 h, under stagnant conditions. TTR proteolysis was stopped by using phenylmethylsulfonyl fluoride (PMSF) at a final concentration of 1.5 mM and, then TTR samples (500 ng/well) were applied into a 15% polyacrylamide SDS-PAGE. After electrophoresis, proteins were stained with PageBlue™ protein staining solution (Thermo Scientific, Waltham, MA, USA) or transferred onto nitrocellulose membrane using iBlot dry blotting system (Thermo Scientific, Waltham, MA, USA). TTR immunoblot was performed using a commercially available antibody produced in our lab, mouse anti-transthyretin mutant (Y78F), clone AD7 (1:100) (Sigma-Merck, Darmstadt, Germany). This monoclonal antibody detects glycosylated form of TTR V30M in plasma and acts as a conformational antibody recognizing specific TTR variants, such as G47A, G49A, S50R, and T59K, in particular conditions [28]. Rabbit polyclonal anti-human TTR (1:1000) (DAKO, Hovedstaden, Denmark) was also used. ECL chemiluminescent reagent (Bio-Rad, Hercules, CA, USA) was used as a detection method using Chemidoc apparatus (Bio-Rad, Hercules, CA, USA). Three independent experiments of in vitro plasmin-mediated proteolysis were performed.

### *4.4. N-Terminal Sequencing Analysis of TTR Fragments*

Plasmin-digested TTR V30M samples (15 μg) were loaded into a 15% polyacrylamide SDS-PAGE gel. Samples were then transferred onto a PVDF membrane (Bio-Rad, Hercules, CA, USA) and proteins were further stained with Coomassie blue R-250 (VWR International, Radnor, PA, USA). Membranes were allowed to dry and the bands below to the TTR monomer, corresponding to TTR fragments (band 1 and band 2) were excised for Nterminal sequencing analysis (Edman degradation method) using an ABI Procise Protein Sequencer, an ABI Microgradient Pump System, and an ABI Programmable Absorbance Detector (Applied Biosystems Inc., Waltham, MA, USA).

#### *4.5. Mass Spectrometry Analysis for the Identification of TTR Fragments*

Gel bands excised from SDS-PAGE were washed twice with 50% acetonitrile (ACN) in 50 mM triethylammonium bicarbonate (TEAB) with shaking at 1500 rpm for 5 min and further treated with ACN twice. Then, proteins were reduced with 25 mM dithiothreitol (DTT) for 20 min at 56 ◦C and alkylated with 55 mM iodoacetamide (IAA) for 20 min at room temperature in the dark, followed by the same wash procedure. Proteins were then digested with trypsin (240 ng) in 50 mM TEAB/0.01% surfactant (ProteaseMAX, Promega, Madison, WI, USA) for 60 min at 50 ◦C. Peptide gel extraction was performed with 2.5% trifluoroacetic acid (TFA) followed by 50% ACN, 0.1% TFA. Samples were dried using Speedvac, resuspended in 10mL 0.1% TFA and cleaned by C18 reverse phase chromatography according to manufacturer's instructions (ZipTip, Sigma-Merck, Darmstadt, Germany).

Sample protein identification and quantification were performed by nano-liquid chromatography mass spectrometry (nano LC-MS/MS), as previously described [50] with a 90 min chromatographic separation run. This equipment was composed of an Ultimate 3000 liquid chromatography system coupled to a Q-Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany). A total of 500 nanograms of each TTR peptide were loaded onto a trapping cartridge (Acclaim PepMap C18 100 Å, 5 mm × 300 μm i.d., 160454, Thermo Scientific, Bremen, Germany) in a mobile phase of 2% ACN, 0.1% FA at 10 μL/min. After 3 min loading, the trap column was switched in-line to a 50 cm × 75 μm inner diameter EASY-Spray column (ES803, PepMap RSLC, C18, 2 μm, Thermo Scientific, Bremen, Germany) at 250 nL/min. The LC separation was achieved by mixing A: 0.1% FA and B: 80% ACN, 0.1% FA with the following gradient: 2 min (2.5% B to 10% B), 50 min (10% B to 35% B), 8 min (35% B to 99% B), and 10 min (hold 99% B). Subsequently, the column was equilibrated with 2.5% B for 17 min. The specific MS parameters were: MS maximum injection time, 100 ms; dd settings: minimum AGC target 7 × 103, intensity threshold 6.4 × 104, and dynamic exclusion 20 s. Data acquisition was controlled by Tune 2.11 software (Thermo Scientific, Bremen, Germany). The UniProt database 2020\_05 for the Homo sapiens proteome (75069 entries) together with a customized TTR amino acid sequence were considered for protein identification. Protein identification was performed with the Proteome Discoverer software v2.5 (Thermo Scientific, Bremen, Germany). Proteins were quantified by Label-Free Quantification—LFQ, with precursor quantification based on intensity.

#### *4.6. Aggregation Studies: Dynamic Light Scattering and Thioflavin T Assay*

DLS measurements were performed at 25 ◦C using Malvern Zetasizer Nano ZS apparatus (Malvern, Worcestershire, UK). Each sample was measured 3 times, and the values exhibited in the curves were the average distributions from those triplicates. Thioflavin T (ThT) assay was performed in PBS (pH = 7.4) using a 96-well black bottom plate. TTR (12.5 μg) and ThT (30 μM) (Sigma-Merck, Darmstadt, Germany) per well were mixed and, the fluorescence was measured at Exc./Em = 450 nm/482 nm using SynergyMx apparatus (BioTeK, Winooski, VT, USA).

### *4.7. Mice*

Mice were kept in a controlled temperature room and maintained under a 12 h light/dark period. Transgenic mice carrying human TTR V30M (HM30) were bred as described before [51]. Both younger (12–13 months; *n* = 12) and older (16–21 months; *n* = 7) HM30 mice were subcutaneously (s.c) injected once a week with 5 mg ASO/kg body weight following previous protocols [48]. Animals were euthanized at week 6 of treatment. Plasma and tissue sections were collected and frozen at −80 ◦C or fixed in formalin for further analysis.

### *4.8. Determination of SerpinA1 and TTR Protein Levels in Mice Heart*

Mouse hearts were homogenized using RIPA lysis and extraction buffer according to the manufacturer's instructions (Santa Cruz Biotechnology, Dallas, TX, USA). Briefly, mouse hearts were homogenized in RIPA buffer using a laboratory homogenizer to disrupt the tissue. Protein homogenates were further frizzed at −80 ◦C to promote cell lysis, further centrifuged at 21,500× *g* for 15 min at 4 ◦C and, the supernatant containing protein was harvested. Protein was quantified in the heart lysates using Bradford protein assay (BioRad, Hercules, CA, USA), and 50 μg of total protein was loaded into a 10% and 15% polyacrylamide SDS-PAGE to evaluate SerpinA1 and TTR expression, respectively. Gels were then transferred onto nitrocellulose membrane using a wet system and, membranes were incubated overnight with rat monoclonal anti-mouse SerpinA1 antibody (1:1000; R&D systems, Minneapolis, MN, USA), rabbit polyclonal anti-human TTR antibody (1:1000, DAKO) or mouse anti-transthyretin mutant (Y78F), clone AD7 (1:100, Sigma-Merck, Darmstadt, Germany). GAPDH was used as a protein loading control, and recombinant rabbit anti-GAPDH antibody was used for immunoblotting (1:100000, Abcam, Cambridge, UK). ECL chemiluminescence reagent (Bio-Rad, Hercules, CA, USA) was used as a detection method using Chemidoc apparatus (Bio-Rad, Hercules, CA, USA). Protein bands were

quantified by densitometry using ImageJ (U. S. National Institutes of Health, Bethesda, MD, USA), and results of protein expression were normalized to GAPDH expression.

#### *4.9. Immunohistochemical Analysis of Tissue TTR Deposition*

Paraffin-embedded sections of both duodenal and cardiac tissue were deparaffinized in xylene and rehydrated in descent alcohol series. Antigen retrieval was performed at 95 ◦C for 15 min using citrate buffer (pH = 6) and, then, endogenous peroxidase activity was quenched in 3% hydrogen peroxide in methanol. Sections were blocked using 10% fetal bovine serum, 1% bovine serum albumin, and 0.5% Triton X-100 in PBS. TTR immunostaining was performed using primary rabbit polyclonal anti-human TTR antibody (1:600, DAKO, Hovedstaden, Denmark) and secondary anti-rabbit antibody (1:200) (Vector, Burlingame, CA, USA). For SerpinA1 staining, primary rat anti-mouse antibody (1:100; R&D systems, Minneapolis, MN, USA) and secondary anti-rat antibody (1:200) (Vector, Burlingame, CA, USA) were used. Tissue slides were developed using 3,3 -diaminobenzidine (DAB) (DAKO, Hovedstaden, Denmark), counterstained with hematoxylin and mounted in Entellan® (Sigma-Merck, Darmstadt, Germany). Images were captured at 10× magnification using Olympus BX50 microscope (Shinjuku, Tokyo, Japan) and analyzed using Image Pro Plus software (Rockville, MD, USA). Results represent the occupied area in pixels corresponding to the substrate reaction color that was further normalized relative to the total image area.

#### *4.10. Protease Activity and Plasmin Activity Fluorescence Measurements*

Plasma samples and heart homogenates from HM30 mice were directly used without any dilution. However, protein from heart homogenates was extracted by using RIPA lysis and extraction buffer without supplementation with protease inhibitors. Protease activity assay and plasmin activity assay kits were used according to the manufacturer's instructions. Briefly, mice samples were incubated with FITC-casein substrate in protease activity assay, whereas in the plasmin activity assay, samples were incubated with a synthetic plasmin AMC-substrate. The fluorescence was measured at Ex/Em = 485/530 nm and Ex/Em = 360/450 nm in protease activity assay kit (Abcam, Cambridge, UK) and plasmin activity assay kit (Abcam, Cambridge, UK), respectively, using SynergyMx apparatus (BioTek, Winooski, VT, USA). Both protease and plasmin activities were calculated according to the manufacturer's instructions.

#### *4.11. Statistical Analysis*

Statistical analysis was performed by one-way ANOVA (Tukey's multiple comparisons as post-test) and unpaired t-test using GraphPad Prism 5 software (San Diego, CA, USA). Statistical significance was considered when *p*-value ≤ 0.05. Results were expressed as mean + standard error of the mean (SEM).

**Supplementary Materials:** Supplementary Materials can be found at https://www.mdpi.com/ article/10.3390/ijms22179488/s1.

**Author Contributions:** Conceived and designed the experiments: F.B., C.N., H.H.J.S., A.Z., M.J.S., M.R.A. Performed the experiments: F.B., P.G. Analyzed the data: F.B., P.G., A.Z., M.J.S., M.R.A. Contributed reagents/materials/analysis tools: S.G., B.P.M., M.J.S., M.R.A. Writing the paper: F.B., A.Z., M.J.S., M.R.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by COMPETE 2020 of PT2020 through the European Regional Development Fund (ERDF), "NETDIAMOND—New Targets in DIAstolic heart failure: from coMOrbidities to persoNalizeD medicine" project financed by the European Structural and Investment Funds (ESIF), through the Programa Operacional Regional (POCI-01-0145-FEDER-016385) and HEALTH-UNORTE: Setting-up biobanks and regenerative medicine strategies to boost research in cardiovascular, musculoskeletal, neurological, oncological, immunological, and infectious diseases, NORTE-01-0145-FEDER-000039. FB was supported by FCT—Fundação para a Ciência e Tecnologia/MEC— Ministério da Educação e Ciência with a PhD fellowship (SFRH/BD/123674/2016).

**Institutional Review Board Statement:** All animal experiments were approved by the Portuguese General Veterinarian Board (authorization number 014982 from DGV-Portugal) and animals were kept and used strictly in accordance with National rules and the European Communities Council Directive (86/609/EEC), for the care and handling of laboratory animals.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data are contained within the article or Supplementary Materials.

**Acknowledgments:** The authors acknowledge technical assistance of Paula Chicau from N-terminal sequencing facility of ITQB Nova, Lisbon, Portugal for N-terminal sequencing experiments and data analysis. The authors also acknowledge to Hugo Osório from Proteomics Unit of i3S (Instituto de Investigação e Inovação em Saúde) for mass spectrometry experiments and analysis.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


**Alfredo Bellon 1,2,\*, Tuna Hasoglu 1, Mallory Peterson 3, Katherine Gao 4, Michael Chen 5, Elisabeta Blandin 6, Alonso Cortez-Resendiz 1, Gary A. Clawson <sup>7</sup> and Liyi Elliot Hong <sup>8</sup>**

	- <sup>3</sup> Department of Engineering Science and Mechanics, Penn State College of Engineering, State College, Philadelphia, PA 19107, USA; mpeterson1@pennstatehealth.psu.edu

**Abstract:** Deficits in neuronal structure are consistently associated with neurodevelopmental illnesses such as autism and schizophrenia. Nonetheless, the inability to access neurons from clinical patients has limited the study of early neurostructural changes directly in patients' cells. This obstacle has been circumvented by differentiating stem cells into neurons, although the most used methodologies are time consuming. Therefore, we recently developed a relatively rapid (~20 days) protocol for transdifferentiating human circulating monocytes into neuronal-like cells. These monocyte-derivedneuronal-like cells (MDNCs) express several genes and proteins considered neuronal markers, such as MAP-2 and PSD-95. In addition, these cells conduct electrical activity. We have also previously shown that the structure of MDNCs is comparable with that of human developing neurons (HDNs) after 5 days in culture. Moreover, the neurostructure of MDNCs responds similarly to that of HDNs when exposed to colchicine and dopamine. In this manuscript, we expanded our characterization of MDNCs to include the expression of 12 neuronal genes, including tau. Following, we compared three different tracing approaches (two semi-automated and one automated) that enable tracing using photographs of live cells. This comparison is imperative for determining which neurite tracing method is more efficient in extracting neurostructural data from MDNCs and thus allowing researchers to take advantage of the faster yield provided by these neuronal-like cells. Surprisingly, it was one of the semi-automated methods that was the fastest, consisting of tracing only the longest primary and the longest secondary neurite. This tracing technique also detected more structural deficits. The only automated method tested, Volocity, detected MDNCs but failed to trace the entire neuritic length. Other advantages and disadvantages of the three tracing approaches are also presented and discussed.

**Keywords:** schizophrenia; autism; stem cells; cytoskeleton; neurite; dendrite; neurodevelopment; biomarker; transdifferentiation and neuronal model

**Citation:** Bellon, A.; Hasoglu, T.; Peterson, M.; Gao, K.; Chen, M.; Blandin, E.; Cortez-Resendiz, A.; Clawson, G.A.; Hong, L.E. Optimization of Neurite Tracing and Further Characterization of Human Monocyte-Derived-Neuronal-like Cells. *Brain Sci.* **2021**, *11*, 1372. https://doi.org/10.3390/ brainsci11111372

Academic Editors: Masaru Tanaka and Lydia Giménez-Llort

Received: 4 September 2021 Accepted: 14 October 2021 Published: 20 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Neurodevelopmental disorders such as autism and schizophrenia are relatively common alignments [1,2] caused by a complex combination of environmental and genetic factors. Unfortunately, treatment and diagnostic methods for these illnesses remain unsatisfactory. It is therefore not surprising that the search for biomarkers is intense [3–8]. However, a crucial step in the development of biomarkers and improvement of treatment as well as diagnosis for any illness is understanding its pathophysiology.

One of the many challenges researchers face when studying neurodevelopmental disorders is that they are diagnosed once most neurodevelopmental stages have been completed. For instance, schizophrenia is diagnosed in late adolescence or early adulthood. Autism is often recognized earlier in life but still too late to study neuronal processes such as neurite formation, neuronal polarization and pruning of neuronal extensions. These neurodevelopmental processes are of particular importance, as the neuronal structure has been consistently associated with the pathophysiology of schizophrenia and autism [9–13]. It is therefore possible that studying early neurostructural rearrangements directly in cells from patients with autism or schizophrenia would lead to a better understanding of its pathophysiology.

An additional challenge when ascertaining neurodevelopmental disorders is the accessibility of neurons coming directly from clinical patients. This obstacle has been circumvented by several different methods. The collection of olfactory neuroepithelial cells (ONCs) is the only approach presently available that provides access to mature neurons [14]. It also delivers glial, epithelial and neuroprogenitor cells as well as neurons at different stages of differentiation [15,16]. In order to access ONCs, a qualified otorhinolaryngologist has to perform a biopsy of the olfactory mucosa [16]. This invasive procedure has limited the use of ONCs. In addition, concerns have been raised about the reproducibility of data when using olfactory mucosa, as biopsies from the same individual can deliver variable results [15,16]. Another approach that circumvents the limited access of neurons coming directly from patients is the use of mesenchymal stem cells (MSCs). MSCs can be rapidly differentiated into neuronal-like cells in vitro [17]. However, the scarce use of MSCs in the study of psychiatric and neurologic disorders appears to be due to difficulties in retrieving MSCs. Obtaining MSCs, often if not always, requires a biopsy [18], which is a surgical procedure that requires consultation with a specialist. There is also another characteristic of MSCs that has determined its fate in research: the fact that MSCs do not trigger an immunological reaction. Such an attribute makes this type of stem cells an excellent tool for cellular transplant [18]. On the other hand, to study neurodevelopmental disorders, the most common stem cells currently used are induced pluripotent stem cells (IPSCs). IPSCs allow researchers to develop different types of neurons with sophisticated neuropils [19]. Even brain organoids that resemble aspects of early brain development can be generated using IPSCs [19]. Unfortunately, generating IPSCs requires altering the cell's genome (reprograming) [20,21], which can become a confounder when studying illnesses with poorly understood genetic predispositions, such as autism and schizophrenia [22]. IPSCs have also been criticized because of difficulties in reproducibility [23,24]. Moreover, the transformation of somatic cell to differentiated neuron is expensive and time consuming [15]. Not surprisingly, published manuscripts involving IPSCs and neurodevelopmental disorders comprise rather small cohorts. Another emerging methodology consists of directly reprogramming somatic cells, often fibroblasts, directly into neurons [25,26]. This approach, known as induced neurons (iNs), bypasses the need for dedifferentiation but still requires altering the cell's genome [25,26]. The potential confounding effects of reprograming and the need to show reproducible results when studying neurodevelopmental disorders remain. However, iNs are becoming a promising alternative for regenerative medicine [27]. A faster approach for obtaining neuronal-like cells that completely avoids genetic reprogramming is transdifferentiation of somatic cells.

We have recently developed a methodology for transdifferentiating human circulating monocytes into neuronal-like cells in only 20 days [28]. These monocyte-derived-neuronallike cells (MDNCs) express several genes and proteins considered neuronal markers. Among the genes and proteins present in MDNCs are NeuN and PSD-95, considered markers for mature neurons. However, MDNCs also express markers of immature neurons such as nestin. Moreover, these cells are not yet committed to developing into any specific neuronal type and instead express markers for glutamatergic, dopaminergic, GABAergic and serotoninergic neurons. Of particular importance for the study of the neuronal structure is the expression of microtubule associated protein 2 (MAP-2) [28], as this protein is a marker for dendrites [29,30]. Tau is another relevant neuronal protein, as it is an axonal marker [29,30]. Immature neuronal extensions that have not yet developed into either axons or dendrites are called neurites [31]. During early stages of neuronal development, MAP-2 is present in all neurites and in the cell soma [29,30]. We have previously shown that MDNCs express MAP-2 in all its extensions as well as in the soma [28]. While expression of tau in MDNCs is still to be proven, the information currently available indicates that MDNCs extend neurites that have not yet developed into either dendrites or axons. However, even at this early stage of neurodevelopment, we have shown that MDNCs conduct electrical activity [28].

In a prior publication, we directly compared the structure of MDNCs with that of human developing neurons (after 5 days in culture) as well as with that of differentiated human neuroblastoma cells [28]. The structure of these three different neuronal cell types was similar [28]. Perhaps more important for the study of schizophrenia and autism is that the structure of MDNCs responds similarly to human neurons and neuroblastoma cells when exposed to dopamine and colchicine [28]. Therefore, MDNCs allow us to study some aspects of the neuronal structure that take place during early development directly in patients' cells that carry the genetic predisposition to illnesses such as schizophrenia and autism. This opens the possibility of starting to unveil the pathophysiology of such neurodevelopmental disorders. While other neurodevelopmental disorders such as attention deficit hyperactivity disorder (ADHD) [32], bipolar disorder [33] and others can also be studied using MDNCs, here we emphasize autism and schizophrenia because deficits in the neuronal structure are consistently found [9–13].

Neurite outgrowth is a key neuronal feature, and therefore characterizing neurites is important for understanding MDNCs' neuronal properties and their application in disease and pharmacology research. Neurites are numerous, and accurate measurements through individual neurite tracing are labor-intensive. Therefore, in order to take advantage of this faster yield of neuronal-like cells, an efficient neurite tracing method is critical for extracting neurostructural data from MDNCs. The current available options can be divided into two general tracing methodologies: automated and semi-automated. Most automated alternatives are similar. They rely on software capable of detecting neurons stained with a fluorochrome. Tagged cells are automatically traced. The output of such softwares is faster than semi-automated methods, as these latter options require the researcher to select the cell to be traced and then identify the beginning and end of the neurite of interest. One of the advantages of semi-automated methods is that researchers have more flexibility when deciding which cells to trace, as they are not bound by the expression of a specific marker. Another advantage is that immunofluorescence or the expression of a fluorescent marker such as green fluorescent protein (GFP) can be avoided. These techniques are not always desirable, as both can lead to cellular damage [34,35].

Postmortem studies indicate that defects in the neuronal structure of patients with neurodevelopmental disorders are subtle [9–13]. Thus, avoiding immunofluorescence or the expression of a fluorescent marker is advantageous, as these techniques could mask inconspicuous defects [34,35]. Another strategy for minimizing structural confounders is using each neuronal-like cell as its own control. For this purpose, cells are identified and photographed before receiving any treatment. Then, after treatment with the compound to be tested and once the desired incubation time has passed, the exact same cells are again identified and photographed. Structural differences in the same cells before and after treatment are reported. We used this approach to determine the structural effects of colchicine and dopamine on MDNCs [28].

In order to avoid the use of fluorochromes and to utilize each neuronal-like cell as its own control, we decided to test three different neurite tracing approaches. First, we traced each MDNC in its entirety using a semi-automated method. Since this approach is lengthy, we also tested a simplified version in which only the longest primary neurite and longest secondary neurite were traced. Finally, the third and most intriguing alternative was to use an automated method without the addition of a fluorochrome. Automated software are triggered by brightness. Neuronal-like cells appear significantly brighter than the background when pictures are taken using light microscopy, and thus, cell recognition is expected.

The first goal of this study was to compare the expression of several previously unreported neuronal markers between MDNCs, human neuroblastoma cells and THP-1 cells (a human monocytic cell line) to further validate MDNCs as neuronal-like cells. The second and main objective was to determine which of the three neurite tracing methods was faster. We hypothesized that an automated method would be faster than the other two semiautomated approaches. The third and final goal was to establish whether any of the three techniques was better at exposing neurostructural defects, with the expectation that whole-cell tracing of MDNCs would reveal the highest number of structural deficiencies, considering the thoroughness of this approach.

### **2. Methods**

### *2.1. Cell Culture*

All blood donors gave their informed and written consent after receiving a full description of the study. Experiments were approved by the Institutional Review Board (IRB) at Penn State University (Study #00006911). Fresh blood was obtained from healthy individuals. We then followed our transdifferentiation protocol, as previously described [28]. Briefly, fresh blood was separated into its components by Ficoll-Paque (17-1440-03, GE Healthcare, Chicago, IL, USA). A fraction of peripheral blood mononuclear cells (PBMCs) was cultured on fibronectin-coated 25 cm2 flasks (13.5 million PBMCs per flask). The remaining PBMCs were used for isolation of CD14+ cells (monocytes) by positive immunomagnetic selection based on the manufacturer protocol (CD14 human microbeads, 130-050-201 Miltenyi Biotec, Auburn, CA, USA). CD14+ cells were cultured on fibronectin-coated wells at a concentration of 180,000 cells per cm2. Plastic plates and flasks came from BD Falcon, Glendale, AZ, USA (351146, 353043 and 353109). Human fibronectin from plasma (F2006, Sigma-Aldrich, St. Louis, MO, USA) was used at a concentration of 20 μg/mL, and coating was carried out overnight at 4 ◦C. Macrophage colony-stimulating factor (MCSF) from AbCys, Paris, France (300-25) was added to monocytes right before culturing at a final concentration of 50 ng/mL (Figure 1A). All cells were maintained in Dulbecco's modified Eagle medium (DMEM), high glucose, GlutaMAX (61965059, GIBCO, Waltham, MA, USA), in which we added 100 U/mL penicillin; 100 mg/mL streptomycin, 1% nonessential amino acids, 1 mM sodium pyruvate, 10 mM HEPES buffer, (all from Life Technologies, Waltham, MA, USA) and supplemented with 10% fetal bovine serum (FBS) Performance Plus from GIBCO (Waltham, MA, USA). Cell culture medium was then replaced on days 4, 7, 10 and 13, as described in Figure 1. The following chemicals and growth factors were added, as shown in Figure 1: butylated hydroxyanisole (BHA) (B1253, Sigma-Aldrich, St. Louis, MO, USA), retinoic acid (RA) (R2625, Sigma-Aldrich, St. Louis, MO, USA), insulin growth factor-1 (IGF-1) (100-11, PeproTech, Cranbury, NJ, USA) and neurotrophin-3 (NT-3) (450-03-100, Peprotech, Cranbury, NJ, USA). On day 17, cell culture media was not replaced; instead, 25 mM potassium chloride (KCL) was added (P5405, Sigma-Aldrich, St. Louis, MO, USA).

**Figure 1.** Transdifferentiation of human circulating monocytes into neuronal-like cells. (**A**) Schematic representation of our 20-day protocol for transdifferentiating human circulating monocytes into neuronal-like cells, starting from day zero (D0) with a blood sample and ending on D20 with neuronal-like cells. Circled arrows represent days on which media were changed. Cell cultured media were replaced with new DMEM, together with PBMCs conditioned media, at a rate of either 2:1 or 1:1 (DMEM/PBMCs), depending on the day of culture, as depicted in the diagram. DMEM was supplemented with different chemicals and growth factors depending on the day of culture, as depicted in the diagram. Exact concentrations are described in the Materials and Methods section. (**B**) Light microscopy photographs of monocytes, right after isolation from PBMCs, monocyte-derived-neuronal-like cells (MDNCs) and human developing neurons (HDNs) in culture for 5 days (20× original magnification). (**C**) Light microscopy photographs of MDNCs in parallel with immunostainings showing tubulin in green and actin in red. The cells' nuclei were stained with DAPI in blue (60× original magnification). Scale bar = 20 μm.

Pictures of cells were taken using a Nikon (Melville, NY, USA) Eclipse Ti-S/L 100 inverted microscope equipped with a CoolSNAP Myo, 20 MHz, 2.8 Megapixel, 4.54 × 4.54 μm pixels camera (Melville, NY, USA) and with a Nikon CFI Super flour 20X DIC prism objective (Melville, NY, USA). Pictures were taken immediately after monocyte extraction and on days 20–21 when transdifferentiation was completed (Figure 1B). Pictures of transdifferentiated cells either under control conditions or after treatment with colchicine 0.5 μM (Sigma-Aldrich, C9754) were identified via a micro-ruled coverslip (Cellattice CLS5-25D, Nexcelom Bioscience, Lawrence, MA, USA). Only neuronal-like cells with at least one primary neurite longer than 2 times the soma size before treatment were traced.

Immunofluorescence was performed as previously described [28] (f. Briefly, after fixation and permeabilization, cells were stained with 4 ,6-diamidino-2-phenylindole, dihydrochloride (DAPI, D1306, Thermo Fisher Scientific, Waltham, MA, USA), mouse antitubulin (1/100, Invitrogen, Waltham, MA, USA), Alexa Fluor-488 (1/200, Life Technology, Waltham, MA, USA) and rhodamine phalloidin (1/200, Invitrogen, Waltham, MA, USA). Images were visualized with a Leica DMI 6000 microscope (Wetzlar, Germany) equipped with a Micro MAX-1300YHS camera using an HCX PL APO 60X oil objective (Leica, Wetzlar, Germany). Images were captured using Metamorph Software (Version 7.1.3, Molecular Devices, San Jose, CA, USA).

### *2.2. Single Cell RNA-Sequencing*

We utilized microfluidic single-cell capture and single-cell mRNA sequencing technologies via Fluidigm's C1TM Single-Cell Autoprep System (C1) to explore genome-wide gene expression in 17 cells exposed to our transdifferentiation protocol and for THP-1 cells. We followed the manufacturer's protocol, as previously described [28]. In short, cells were loaded using an integrated fluidic circuit (IFC) chip that allowed capturing a single cell per well. After optical confirmation of cell number at each capture site on the chip, the cells were processed for in-line cell lysis, reverse transcription and cDNA amplification steps. The resulting cDNA was converted to a sequencing library using Illumina's Nextera XT library preparation kit. The *Rapid* mode of Illumina HiSeq 2500 was used to generate sequencing reads of sufficient depth (about 3 million of sequencing reads) per each cell. Demultiplexed sequencing reads passed the default quality filtering of the Illumina CASAVA pipeline (v1.8, Ilumina, Inc., San Diego CA, USA) and were then exposed to further quality trimming/filtering using FASTX-Toolkit (v.0.0.13, Hannon Laboratory, Cold Spring Harbor, NY, USA). The filtered reads were aligned to the most recent reference genome (hg38) using Tophat (v2.0.9, Center for Computational Biology, Baltimore, MD, USA) [36] by allowing up to 2 mismatches. After normalization was performed via the median of the geometric means of fragment counts across all libraries, fragments per kilobase per million (FPKM) mapped reads values were calculated using Cuffdiff tool, which is available in Cufflinks version 2.2.1 (Trapnell Lab, Seattle, WA, USA) [37]. Some results from this experiment were reported previously [28], but the expression of all genes presented in this manuscript have never before been reported in MDNCs.

Gene expression for human neuroblastoma cells was obtained from a public database generated by Li et al. [38].

#### *2.3. Statistical Analysis*

The non-parametric Mann–Whitney test was used to make pairwise comparisons between MDNCs treated with colchicine versus MDNCs under control conditions. A oneway ANOVA followed by Bonferroni correction was used to make comparisons between the time it took to trace MDNCs using each of the three tracing methods tested. *p* values lower or equal to 0.05 were considered significant.

#### **3. Results**

### *3.1. Neuronal and Monocyte Markers in MDNCs, SH-SY5Y and THP-1 Cells*

We compared the expression of 12 neuronal markers between (a) MDNCs; (b) SH-SY5Y cells, a human neuroblastoma cell line commonly used to study neuronal processes; and (c) THP-1 cells, a human monocytic cell line (to serve as negative control). Of the 12 neuronal markers, 7 are involved in synaptic functions, 4 are part of the neuronal structure and one is a gamma-aminobutyric acid (GABA) type A receptor (Table 1). The expression of these 12 genes has never been reported in MDNCs. Expression of these 12 genes in 17 MDNCs was determined by single-cell mRNA sequencing. All 12 neuronal markers were expressed in at least one MDNC, and most genes were expressed in at least 6 MDNCs (Table 2). SH-SY5Y cells also expressed all of these neuronal genes, while they were not expressed in THP-1 cells (Table 2). We then tested whether two markers for monocytes were present in THP-1 cells, MDNCs or SH-SY5Y. As expected, these two monocyte-specific genes were highly expressed by THP-1 cells, whereas they were not expressed by undifferentiated neuroblastoma cells and were barely detectable in differentiated SH-SY5Y cells (Table 2). Only 1 out of the 17 MDNCs showed very low expression of 1 monocyte marker, and none were expressed in the remaining 16 MDNCs (Table 2).


**Table 1.** Neuronal and monocytic genes and their functions.

**Table 2.** Expression of 12 neuronal markers and 2 markers for monocytes in THP-1 monocytic cells, SH-SY5Y neuroblastoma cells and 17 MDNCs.


1 \* undifferentiated SH-SY5Y; 2 \* differentiated SH-SY5Y; data from Li et al. 2015 [38].

#### *3.2. Whole-Cell Tracing*

After 20 days in culture following our protocol [28], transdifferentiated monocytes acquired a neuronal morphology comparable with that of HDNs (Figure 1B). These MD-NCs extended neurites with a microtubule-based shaft, as shown in Figure 1C. Colchicine is well-known for its ability to elicit neurite retraction [53,54] via microtubule depolymerization [55]. In a prior publication, we showed that the structure of MDNCs responds similarly to the structure of neuroblastoma cells and that of human neurons in vitro when treated with colchicine 0.5 μM [28]. While retraction is expected with colchicine 0.5 μM, minimal to no retraction should occur under control conditions. To determine whether MDNCs exhibited any retraction under control culture conditions, a group of MDNCs was identified and photographed at baseline, meaning at time zero (T0 h). These MDNCs were kept under control conditions for 1 h (T1 h), and pictures of the exact same MDNCs were taken again (Figure 2A). The same procedure was followed to establish whether colchicine elicited pruning of neuronal extensions. In this latter case the 1 h incubation period was carried out in the presence of colchicine 0.5 μM (Figure 2A).

**Figure 2.** Whole-cell tracing of MDNCs after treatment with colchicine 0.5 μM. (**A**) Light microscopy photographs of the exact same MDNCs before (T0 h) and after one hour (T1 h) under control conditions or after treatment with colchicine 0.5 μM (20× original magnification). Scale bar = 20 μm. (**B**) Bar graphs comparing MDNCs' structural response to colchicine versus MDNCs under control conditions. Structural parameters include longest primary neurite (LPN), longest secondary neurite (LSN), number of primary neurites, number of secondary neurites, number of tertiary neurites and total number of neurites. Data are presented as mean ± SEM. Differences were assessed using the non-parametric Mann–Whitney test. For LPN, number of primary, number of secondary, number of tertiary and total number of neurites, *n* = 96 for control and *n* = 82 for colchicine. For LSN, *n* = 91 for control and *n* = 75 for colchicine. \* *p* = or < 0.05.

The principal investigator (PI), who has ample experience tracing cells, traced these four sets of MDNCs, meaning cells that were cultured under control conditions at T0 h and T1 h, as well as cells treated with colchicine at T0 h and T1 h. Since the entire neuropil of each MDNC was traced, we can report differences in the longest primary neurite (LPN), longest secondary neurite (LSN), number of primary neurites, number of secondary neurites, number of tertiary neurites and total number of neurites. MDNCs were traced using a semi-automated software called FIJI (NIH, Bethesda, MD, USA), which is a plugin for ImageJ, an open source image processing program provided by the National Institutes of Health (NIH).

To determine whether there was retraction of LPN or LSN, the percentage of neurite remaining at T1 h was calculated (T1 h/T0 h) for MDNCs cultured under control conditions, as well as cells treated with colchicine 0.5 μM. Then, non-parametric statistical analyses were conducted to establish whether there were differences between control (CTL) and colchicine (Colchi). Whole-cell tracing by the PI evidenced a statistically significant reduction in the percentage of LPN after treatment with colchicine 0.5 μM (CTL, 99 ± 2%; Colchi, 86 ± 2%; *p* = 0.0006), while there were no differences in LSN (CTL, 120 ± 8%; Colchi, 110 ± 6%; *p* = 0.22) (Figure 2B). To establish differences in the number of neurites pruned, we subtracted the number of neurites at T0 h from the number of neurites at T1 h for MD-NCs cultured under control conditions as well as for cells treated with colchicine 0.5 μM. Whole-cell tracing did not reveal differences in the number of primary (CTL, 0.14 ± 0.12; Colchi, 0.18 ± 0.16; non-parametric analysis *p* = 0.48), secondary (CTL, 1.06 ± 0.4; Colchi, 1.2 ± 0.4; *p* = 0.45), tertiary (CTL, 0.010 ± 0.11; Colchi, 0.073 ± 0.14; *p* = 0.52) or total neurites pruned (CTL, 1.25 ± 0.43; Colchi, 1.46 ± 0.45; *p* = 0.27) (Figure 2B).

#### *3.3. Three Neurite Tracing Approaches*

Three neurite tracing approaches were tested by four individuals with research experience (two medical students with previous research experience, one neuroscience graduate student and one laboratory technician). One of the medical students had performed cell tracing before participating in this study, whereas all other participants had no experience in tracing. The tracers were blinded to the treatment condition each of the two groups of MDNCs had received (either CTL or Colchi), and they were unaware of that we were expecting pruning of neuronal extensions with Colchi. Participants were trained on how to use FIJI (NIH, Bethesda, MD, USA), a semi-automated software, and Volocity (Quorum Technologies, Ontario, Canada), an automated software. After training was completed, participants were told to use FIJI for whole-cell tracing, and they were instructed to trace all neurites present in MDNCs for this first tracing method. For the second tracing method, they were again instructed to use FIJI, but this time to trace only the longest primary and longest secondary neurite in each of the MDNCs. This second approach was named the longest neurite method (LN). For the third method, participants processed the photographs of MDNCs for each of the two treatment conditions through Volocity and then confirmed that the MDNCs had been traced. Finally, participants were asked to record the time it took them to trace MDNCs with each of the three tracing methods.

Volocity automatically traces the length and width of the cell, which in the case of neuronal-like cells approximates to the longest primary neurite and the longest secondary neurite. Since two of the three tracing methods tested only provided information on LPN and LSN, we compared the three tracing approaches based on these two neuronal extensions. None of the four individuals encountered any statistically significant difference in LPN after whole-cell tracing (participant 1 (P1), CTL, 95 ± 2%; Colchi, 100 ± 3%; *p* = 0.35; P2, CTL, 100 ± 5%; Colchi, 100 ± 2%; *p* = 0.86; P3, CTL, 100 ± 2%; Colchi, 99 ± 2%, *p* = 0.44; P4, CTL, 99.9 ± 2.8%; Colchi, 100 ± 4.6%; *p* = 0.45) (Figure 3A). Participant 3, however, found colchicine elicited a statistically significant retraction of LSN (P3, CTL, 118 ± 10.3%; Colchi, 88.1 ± 9.7%; *p* = 0.004), while all other participants observed no differences (P1, CTL, 97 ± 12.4%; Colchi, 109.9 ± 11.3%; *p* = 0.24; P2, CTL, 100.6 ± 9.5%; Colchi, 88.9 ± 6.2%; *p* = 0.27; P4, CTL, 104.5 ± 10%; Colchi, 100.1 ± 7.5%; *p* = 0.71) (Figure 3A).

When tracing only longest neurites, one of the participants found colchicine elicited a statistically significant retraction of LPN when compared with MDNCs under control conditions (P2, CTL, 100 ± 4%; Colchi, 90 ± 2%; *p* = 0.04) (Figure 3B). All other participants found no statistical differences in LPN (P1, CTL, 93 ± 4%; Colchi, 92 ± 3%; *p* = 0.66; P3, CTL, 95 ± 4%; Colchi, 100 ± 3%; *p* = 0.34; P4, CTL, 91.7 ± 2.7%; Colchi, 91.4 ± 3.4%; *p* = 0.42). The same participant who found a significant retraction of LSN while tracing the entire cell again encountered retraction elicited by colchicine while tracing only longest neurites (P3, CTL, 143.2 ± 15.8%; Colchi, 84.6 ± 7%; *p* = 0.001) (Figure 3B). Two other participants found no statistical differences in LSN (P1, CTL, 100 ± 14.9%; Colchi, 119.7 ± 16.3%; *p* = 0.07; P2, CTL, 100.7 ± 7.8%; Colchi, 116.2 ± 17.1%; *p* = 1.0), and one participant did not trace LSN (Figure 3B).

The use of Volocity rendered no statistically significant differences in LPN for any of the participants (P1, CTL, 90.2 ± 5.5%; Colchi, 81.4 ± 11%; *p* = 0.14; P2, CTL, 94.6 ± 9.6%; Colchi, 120.7 ± 17.4%; *p* = 0.18; P3, CTL, 95 ± 6.4%; Colchi, 91.7 ± 5.7%, *p* = 0.68; P4, CTL, 116.6 ± 18.2%; Colchi, 92.2 ± 11.8%; *p* = 0.27) (Figure 3C). However, participant 4 found a significant retraction in LSN after treatment with colchicine (P4, CTL, 156.2 ± 18.4%; Colchi, 117 ± 20.3%; *p* = 0.05) (Figure 3C). None of the other participants found statistical differences in LSN while using Volocity (P1, CTL, 202.5 ± 25.5%; Colchi, 182.9 ± 33.6%; *p* = 0.27; P2, CTL, 148.6 ± 16.1%; Colchi, 173 ± 38.5%; *p* = 0.99; P3, CTL, 121.2 ± 12%; Colchi, 119.4 ± 16.4%; *p* = 0.31) (Figure 3C).

**Figure 3.** Comparison between three different tracing methods. (**A**) Bar graphs comparing MDNCs' structural response to colchicine versus MDNCs under control conditions after tracing each MDNC in its entire. Structural parameters include longest primary neurite (LPN) and longest secondary neurite (LSN). Data are presented as mean ± SEM. Differences were assessed using the non-parametric Mann–Whitney test. Participant 1 (P1): for LPN *n* = 37 for control and *n* = 32 for colchicine; for LSN, *n* = 23 for control and *n* = 20 for colchicine. P2: for LPN *n* = 38 for control and *n* = 35 for colchicine; for LSN, *n* = 15 for control and *n* = 19 for colchicine. P3: for LPN *n* = 47 for control and *n* = 34 for colchicine; for LSN, *n* = 41 for control and *n* = 29 for colchicine. P4: for LPN *n* = 51 for control and *n* = 30 for colchicine; for LSN, *n* = 31 for control and *n* = 15 for colchicine. (**B**) Bar graphs comparing MDNCs' structural response to colchicine versus MDNCs under control conditions after only tracing LPN and LSN. The statistical assessment and data presentation are the same as in (**A**). P1: for LPN *n* = 39 for control and *n* = 36 for colchicine; for LSN, *n* = 22 for control and *n* = 17 for colchicine. P2: for LPN *n* = 38 for control and *n* = 36 for colchicine; for LSN, *n* = 21 for control and *n* = 17 for colchicine. P3: for LPN *n* = 30 for control and *n* = 35 for colchicine; for LSN, *n* = 28 for control and *n* = 34 for colchicine. P4: for LPN *n* = 42 for control and *n* = 37 for colchicine; P4 did not trace LSN. (**C**) Bar graphs comparing MDNCs' structural response to colchicine versus MDNCs under control conditions after tracing MDNCs' length and width using the automated tracing software Volocity. The statistical assessment and data presentation are the same as in (**A**). P1: for LPN *n* = 30 for control and *n* = 26 for colchicine; for LSN, *n* = 30 for control and *n* = 23 for colchicine. P2: for LPN *n* = 25 for control and *n* = 18 for colchicine; for LSN, *n* = 25 for control and *n* = 18 for colchicine. P3: for LPN *n* = 33 for control and *n* = 30 for colchicine; for LSN, *n* = 33 for control and *n* = 30 for colchicine. P4: for LPN *n* = 27 for control and *n* = 24 for colchicine; for LSN, *n* = 27 for control and *n* = 24 for colchicine. \* *p* = or < 0.05. (**D**) Dot plot comparing the time in minutes necessary for completing the tracing of all MDNCs (control + colchicine) with each of the three tracing methods: whole cell, longest neurite and Volocity. Data are presented as mean ± SEM. Differences were assessed using a one-way ANOVA followed by Bonferroni correction. For whole cell, longest neurite and Volocity *n* = 8. \* *p* < 0.005, \*\* *p* < 0.002 and \*\*\* *p* < 0.00002.

A one-way ANOVA revealed that the amount of time necessary to complete all tracings was significantly different between each of the three approaches (F(2, 21) = 25.74, *p* < 0.00001) (Figure 3D). Bonferroni correction indicated that tracing longest neurites (LN) took less than half of the time needed to trace the whole cell (WC) (LN, 77.8 ± 8.5 min; WC, 153.6 ± 17.9 min; *p* = 0.001), while tracing the entire cell was more efficient than using Volocity (WC, 153.6 ± 17.9 min; V, 332.5 ± 39.9 min; *p* = 0.001) (Figure 3D). Since one of the students did not trace longest secondary neurites when applying the LN approach, we ran another one-way ANOVA excluding that individual's LN data. The results remained significant (F(2, 19) = 20.81, *p* = 0.00001).

### **4. Discussion**

We have previously shown that MDNCs conduct electrical activity and express a wide variety of neuronal markers [28]. Here we expanded the list to include 12 neuronal genes: 7 involved in synaptic transmission [39–45], 4 associated with neuronal structure [46–49] and 1 gamma-aminobutyric acid (GABA) receptor [50] (Tables 1 and 2). Several of these genes are implicated in the pathophysiology of neurodevelopmental illnesses. For instance, neurexin 3 has been linked to autism [56], whereas SV2A and VAMP are associated with schizophrenia [57,58]. Another synaptic gene, SNAP-25, has been implicated in the etiology of both illnesses [59,60]. Tau and GAP-43 are essential for the development of neuronal structure [47,49]. While tau is commonly known for its association with Alzheimer's disease, this protein has also been linked to schizophrenia [61]. Similar to tau, GAP-43 is crucial for outgrowth of neuronal extensions [49], and not surprisingly, abnormalities in the expression of GAP-43 have been associated with both schizophrenia [62] and autism [63]. Other proteins relevant for the establishment of neuronal shape during development and often involved in the pathophysiology of schizophrenia, such as MAP-2 [64,65], are also expressed by MDNCs [28]. At the same time, markers for monocytes such as CD11B and CCR2 [51,52] are no longer present in MDNCs (Tables 1 and 2).

Several lines of evidence strongly indicate that deficits in the neuronal structure are implicated in the pathophysiology of autism and schizophrenia [12,13,66–68]. However, the inaccessibility of neurons coming directly from living patients' brains has limited the study of early neurodevelopmental processes that transform neuronal structure. MDNCs not only express a variety of genes crucial in sculpting neuronal shape, but in addition, the structure of MDNCs is comparable with that of human neurons after 5 days in culture and also with that of differentiated human neuroblastoma cells [28]. Moreover, the structure of MDNCs responds similarly to that of neurons and neuroblastoma cells when treated with dopamine and colchicine [28].

MDNCs' ability to reproduce characteristics of the structure of human neurons opens the opportunity for studying these aspects of neurodevelopmental illnesses directly in living patients' cells. This means that MDNCs provide a window into early neurodevelopmental processes in vitro, even when patients are already adults. Nonetheless, in order to maximize the delivery of neurostructural results, it is imperative to determine which neurite tracing method is more efficient in extracting data from MDNCs.

Unfortunately, there is no universal tracing method that can efficiently extract neurostructural data under all research conditions. Instead, experts recommend testing several tracing approaches to determine which is the best suited for each laboratory [69,70]. Currently, there is a plethora of automated tracing methods, but the gold standard continues to be manual tracing via semi-automated approaches [69]. Therefore, here we tested three different tracing approaches: (1) whole-cell tracing, (2) longest neurite tracing and (3) Volocity. The first two are semi-automated and thus require more work, while the third method is completely automated. However, before comparing these three tracing methods, we had to establish the right conditions for comparison. Therefore, the principal investigator, who has ample experience tracing cells, traced two separate groups of MDNCs: one cultured under control conditions and one treated with colchicine 0.5 μM. This compound is well-known for its capacity to cause neurite retraction via microtubules depolymerization [55]. Furthermore, we have previously shown that colchicine elicits pruning of neuronal extensions in MDNCs in a way similar to what is found in neurons [53] and neuroblastoma cells [54].

Using the more thorough tracing approach—namely, whole-cell tracing—the PI found that colchicine elicited, as expected, a statistically significant retraction of LPN (Figure 2B). None of the other structural parameters revealed statistical differences (Figure 2B). Then, four other individuals, mostly students with research backgrounds, traced the same two groups of MDNCs (control versus colchicine 0.5 μM) using the three different tracing methods. These four participants were blinded to the treatment condition they were tracing. All the statistically significant retractions found by these four participants were, as expected, caused by colchicine (Figure 3A–C). However, relatively few structural differences were found. This is not entirely surprising, as these four individuals had limited experience with tracing, and the differences between the two MDNCs groups were subtle (Figure 2B). It is important to note that the tracing approach that yielded more statistically significant findings was the simplest of all, meaning the approach that only traced the longest primary and longest secondary neurite (Figure 3B).

The most surprising finding was that Volocity, the automated tracing method, was the slowest in delivering structural results (Figure 3D). This delay was not due to lack of recognition of MDNCs, even though these cells were not marked with a fluorochrome. Instead, pictures of MDNCs were live. The difficulties arose because in many instances Volocity did not identify the entire neuritic length. Students, therefore, had to piece together sections of neurites, similar to what other research teams have described using different automated softwares [70]. This task was more time consuming than even tracing the entire cell using a semi-automated approach (Figure 3D). Having MDNCs stained with a fluorochrome would have eliminated the need for reconstruction of the neuritic length. However, given the inherent damage attached to cell fixation and permeabilization [34,35], it is questionable whether the subtle structural differences between control MDNCs and those treated with colchicine would have been observed.

The fastest tracing approach was tracing only the longest primary and longest secondary neurites (Figure 3D). This approach was also the one that yielded more statistically significant differences between treatment conditions (control versus colchicine 0.5 μM) (Figure 3B). We were expecting whole-cell tracing to detect more structural differences, given the precision of this method. However, perhaps the simplicity of tracing only two neurites per MDNC as opposed to delineating the entire neuropil improves accuracy. Given that tracing longest neurites was more accurate and took half the time as tracing the entire cell and a quarter of the time as Volocity (Figure 3D), we recommend this tracing approach for future studies on the structure of MDNCs.

There are other factors that need to be considered when selecting a tracing approach. One is that tracing only the longest neurites neglects other structural parameters. Additionally, Volocity is just one of the many commercially available automated software products. It is possible that other automated applications would recognize the entire structure of MDNCs, even when analyzing photographs of live cells using light microscopy. However, if automated software products become a viable alternative, cost will have to be factored in, as semi-automated methods usually do not bear any cost to the researcher, while most automated software have to be purchased [69,70].

In summary, selecting a tracing method is a complicated process that depends on the specific research conditions to be tested [69,70]. For instance, analyzing neurons in culture (2-dimensional) versus brain slices (3-dimensional) or studying intact neurites versus damaged neurites would each generate its own set of intricacies for which only a couple of tracing approaches would be suitable. It is also essential to determine which aspects of the neuronal structure will be studied, as some tracing paradigms are better at measuring neurite length, while others excel at counting number of extensions [69,70]. Therefore, experts recommend testing different tracing approaches to determine the most efficient method for each laboratory [69,70]. Here, we determined that the most efficient tracing strategy for studying neuritic length in MDNCs is tracing only the longest primary

and longest secondary neurites (Figure 3D). The limitation of this modality is that it does not provide information about the number of neurites or other aspects of the neuropil, such as number or length of tertiary or quaternary neurites. Another limitation of our study is that we only conducted tracing using one automated method among the many currently available [69,70]. Future studies will have to be conducted to determine whether other automated paradigms prove better at extracting neurostructural data from MDNCs.

#### **5. Conclusions**

MDNCs express a wide variety of neuronal markers that have been associated with the pathophysiology of autism and schizophrenia. Since MDNCs originate from a blood sample taken directly from patients, these cells carry the genetic susceptibility to the neurodevelopmental illness that the patients are afflicted with. In contrast with rodent neurons in culture or neuronal cell lines such as neuroblastoma cells, MDNCs allow researchers to study directly in patients' cells early neurodevelopmental processes involving changes in neuronal structure. In order to maximize efficiency in studying MDNCs' structure, the best approach is to only trace the longest primary neurite and the longest secondary neurite using FIJI, a semi-automated software made available by the NIH.

**Author Contributions:** A.B. envisioned and designed this project, wrote the first draft and conducted some of the experiments; T.H. helped with result analysis; M.P. helped with the design of this project; K.G. conducted some experiments and helped write the final draft; M.C., E.B. and A.C.-R. conducted some experiments; G.A.C. conducted some experiments and helped write the final draft; L.E.H. helped design the project and helped write the final draft. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and the study was approved by the Institutional Review Board of Penn State Hershey Medical Center (STUDY00006911).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Acknowledgments:** The authors would like to thank the Ling and Esther Tan Early Career Professorship endowment given to A.B. We also would like to thank Julia Lesperance for her technical assistance.

**Conflicts of Interest:** This protocol is patented in the USA (99932556 (B2)) and Europe (2862926 (A1 & B1)). This patent is held by A.B. in collaboration with other authors as well as INSERM and SATT IDF-Innov. The authors report no other financial conflict of interest related to this manuscript.

### **References**

