Introduction

The primary cilium is a microtubule-based sensory organelle that extends from the plasma membrane. It is present in all cell types and plays a vital role in physiological and developmental processes by controlling different signalling pathways such as WNT, Sonic hedgehog (SHh), and transforming growth factor β (TGF-β)1,2,3,4,5. Ciliary dysfunction generates different phenotypes, the most common of which are obesity, type 2 diabetes mellitus, retinopathies or the appearance of multi-organ fibrosis leading to liver, kidney, or lung dysfunction5,6,7,8.

The group of diseases characterised by ciliary dysfunction, mutations in primary cilium or basal body genes, are commonly known as ciliopathies. Alström syndrome (ALMS) and Bardet-Biedl syndrome (BBS) are two model ciliopathies where, despite the progress made in recent years, a clear pathological mechanism has not yet been established9,10,11,12. ALMS is a monogenic recessive disease with an incidence of 1 in 1,000,000 individuals in the general population. To date, the only known causative gene is the ALMS1 gene, which codes for a centrosome-associated protein, responsible for maintaining cohesion between the mother centriole and the daughter centriole13,14,15. On the other hand, BBS is also an autosomal recessive disorder, with 23 causative genes described to date, named from BBS1 to BBS23, being BBS1 the most prevalent gene with pathogenic variants in 23% of described cases10,11,12. The main proteins of BBS, BBS1, BBS2, BBS4, BBS5, BBS7, BBS8, BBS9 and BBS18/BBSIP1, encode for an octameric protein complex called the BBSome involved in primary cilia transport (IFT), retrograde and anterograde16,17. The clinical phenotypes of these two ciliopathies are overlapping, making their differentiation difficult and complicating the correct diagnosis18. One of the main causes of death in adult patients with these syndromes is chronic kidney disease (CDK)13,19,20,21,22,23,24. This disease is usually accompanied by inflammatory and fibrotic processes that progressively degrade normal kidney function until eventually renal failure and death occur25.

The TGF-β pathway is one of the main pathways involved in fibrotic and inflammatory processes at the cellular level26,27. It is related to the control of several processes such as mitosis, apoptosis, cell migration or autophagy through cross-signalling with other pathways such as MAPK, PI3K/AKT or SMADs28,29,30,31,32. In previous studies in our laboratory, we have determined that ALMS1 depletion leads to altered signal transduction of the TGF-β pathway and other related pathways such as AKT or p5333,34,35, altering processes such as cell migration or EMT.

In this study, we apply an original approach based on phosphoproteomics to delve into the regulation of the TGF-β pathway in ALMS and to determine for the first time whether there are alterations in this pathway in BBS.

Materials and methods

Cell culture and CRISPR knockouts

The hTERT-BJ-5ta cell line, provided by the American Type Culture Collection (ATCC), was maintained in a medium composed of 4:1 of Dulbecco’s minimal essential medium (DMEM, Gibco, Invitrogen, NY, USA) and Medium 199, Earle’s Salts (Gibco, Invitrogen, NY, USA), supplemented with 10% fetal bovine serum (FBS) (Gibco, Invitrogen, NY, USA) and 2% penicillin/streptomycin (P/S) (Gibco, Invitrogen, NY, USA). In this study, a knockout model for the BBS1 gene was generated (Figure S1 A, B) using 4 different sgRNAs (Supplementary Table S1-2) and a knockout (KO) model for the ALMS1 gene previously generated in Bea-Mascato et al.34 was used. The methodology used in the generation of the BBS1 model was the same as described in Bea-Mascato et al.34. The sgRNAs for BBS1 were designed with the Benchling tool (CA, USA) and a minimum score of 75.3 was used for off-targets to prioritise specificity (Supplementary Table S1). BJ-5ta were transduced with lentiviral particles with the protocol described in Bea-Mascato et al.34. The BBS1 KO model was validated at the genomic level by PCR and at the protein level by Western blot using the anti-BBS1 antibody (Abcam, ab166613).

As previously reported, we validated the ALMS1 KO at both the genomic and transcript levels. Genomic confirmation was achieved by PCR, which verified the 113-bp deletion and the consequent premature stop codon, while transcript validation was carried out by quantitative PCR (qPCR). Protein-level verification was not pursued because no antibody currently provides reliable detection of ALMS1.

TGF-β stimulation and protein extraction

Cell lines were cultured in 150 mm dishes (Corning, NY, USA) until 90% confluence was reached and then serum starved for 24 h. The following day, rhTGF-β1 ligand (R&D Systems; 240-B) was added to the plates at a final well concentration of 2ng/mL for 0 and 30 min. The plates were then washed 3 times with PBS and the cells were harvested in a volume of 1.5mL of PBS. They were then pelleted at 10,000 rpm for 5 min in a Sigma® 1–14 K at 4 °C. The PBS volume was removed from the tubes and the pellets were frozen at − 80°C until the experiment was continued.

Protein extraction was performed by adding 500µL of TEAB 100mM lysis buffer with 1% SDS (Thermo Fisher, Waltham, USA) to each sample. Samples were sonicated by giving five 5-s pulse intervals with a wave amplitude of 20% on a Branson ultra sonicator model 102C (Branson Ultrasonic, Connecticut, USA). They were then left to stand on ice for 10 min with moderate agitation. Finally, the cell debris was removed by centrifuging the samples for 30 min at 12000rpms in a Sigma® 1–14 K at 4°C. The supernatant from the tubes was transferred to low-binding tubes (Thermo Fisher, Waltham, USA) and stored at − 80 degrees until the next step of the analysis. Quantification of protein lysates was performed using the Pierce™ BCA Protein Assay Kit (Thermo Fisher, Waltham, USA).

Reduction, alkylation and precipitation.

Initially, samples were centrifuged at 16,000 rpm for 10 min at 4°C. Then, 600µg per sample was transferred into a new 1.5 mL tube and adjusted to a final volume of 600µL with 100mM TEAB buffer. Samples were divided into several tubes (200µL/tube). 10µL of 200mM TCEP was added to each tube and incubated at 55 °C for 1 h. Then, 10µL of 375mM iodoacetamide (IAA) was also added to the sample and incubated for 30 min protected from light at room temperature (RT). Finally, 6 volumes (1200µL) of acetone pre-cooled to − 20 degrees were added and left to precipitate overnight. The next day the samples were centrifuged at 8000rpms for 12 min at 4 °C, the acetone was decanted off and the pellet was left to dry for 2–3 min.

Protein digestion and cleaning, TiO2 enrichment, TMT labelling and LC–MS/MS analysis

Approximately 200µg of protein pellets precipitated in acetone were resuspended with 200 µL of 100mM TEAB buffer. Then 5µL (5µg) of trypsin was added per tube and digested at 37 °C overnight. The next day, 10 µL of 5% trifluoroacetic acid (TFA) was added to reduce the pH of the solution. Finally, peptide samples were cleaned up using C18 desalting tips (Agilent OMIX 100 µL, SPE with pipette) and the pooled eluents with speed-vac at 45 °C. Phosphopeptide enrichment was performed using TiO2 spin tips from the High-SelectTM TiO2 Phosphopeptide Enrichment Kit following the manufacturer’s protocol (Thermo Fisher, Waltham, USA). After enrichment, a further cleaning was performed using C18 desalting tips.

Samples were resuspended in 50 µL of 100mM TEAB buffer and 20 µL of TMT label reagent was added to each sample, using TMT10plex™ Isobaric Mass Tagging Kit (Thermo Fisher, Waltham, USA). Samples were multiplexed and analyzed by LC–MS/MS using a 90-min gradient on Orbitrap Eclipse (Thermo Fisher, Waltham, USA) with a method using CID-MSA for the identification of MS2 and a real-time search algorithm to quantify reporters in MS3. As a quality control, the BSA controls were digested in parallel and ran between each of their samples to avoid carryover and assess instrument performance.

Protein identification and differential phosphorylation analysis

The identification of peptides/proteins was performed using the software Proteome Discoverer software (v2.5) (Supplementary Table S3-4). Samples were searched against the SP_Human database (March 2021), using the search algorithm Mascot v2.6 (http://www.matrixscience.com/). Peptides have been filtered based on false discovery rate (FDR) and only peptides showing an FDR lower than 5% have been retained. The peptides were grouped into proteins with the default option of Proteome discovery (v2.5) and these were used in the downstream analysis.

The proteins resulting from the identification were used for differential phosphorylation analysis. Each multiplexed/batch (with and without stimulation of the TGF-β pathway) was analysed individually with the R (v4.2.1) package DEP36. The data was filtered to keep only the proteins that were quantified in all samples of at least one condition. Subsequently, the normalization of the data by variance stabilizing normalization (VSN) was performed. Finally, for data imputation, the MiniProb algorithm was used with a q-value of 0.01. The log2 fold change (FC) of the differentially phosphorylated proteins was calculated against the controls (WT) of the corresponding batch for both knockouts (ALMS1 and BBS1) (Supplementary Table S5-6). For downstream analyses, the differentially phosphorylated protein was that with a value of |log2FC|> 0.5 and an FDR < 0.05.

Network diffusion, enrichment analysis and protein–protein interaction analysis

For network diffusion analysis, a graph covering the entire human signalome was created with the R package, igraph37. The creation of the human interactome was performed as follows. Initially, the human interactome was extracted from the IntAct database (version: 4.2.17, last updated May 2021)38. In addition, kinase-substrate interactions and kinase-kinase interactions contained in PhosphoSitePlus38 (version 6.5.9.3, last update May 2021), OmniPath39 (last version May 2021) and SIGNOR 2.040 (last version May 2021) were integrated in this interactome. All proteins not annotated in Swiss-Prot (UniProt, 2021) and those not annotated with at least one GO term (Gene Ontology, 2021) were removed. The resulting protein interaction network (PIN) comprised a total of 16,407 nodes and 238,035 edges. The weight of edges was defined by the semantic similarity between gene pairs, modelled according to Topological Clustering Semantic Similarity (TCSS) algorithm41 and calculated using the Semantic Measure Library42. The raw PIN was normalised to avoid hub bias using “correct.for.hubs” option in diffusR.

The differentially phosphorylated proteins detected in the “Protein identification and differential phosphorylation analysis” section, were used as seed nodes in the interactome graph. These nodes were assigned the same weight and a weight of 0 to the rest of the network. Finally, the random walk with restart (RWR) algorithm43,44 was used through the R package diffusR using a restart probability of 0.3, 10,000 iterations and a threshold of 1e-06. Nodes were sorted by RWR score after diffusion, and to ensure at least 10% real (not inferred) data, the top 100 were used in subsequent analyses (Supplementary Table S7-8).

The enrichment analysis was performed using the R package enrichR, an R interface for accessing the enrichR database45. The differentially phosphorylated proteins and the 100 nodes with the highest RWR scores were compared with the GO Molecular Function 2021, GO Biological Process 2021 and BioPlanet 2019 datasets in independent enrichment analysis. Only the 20 processes with the lowest FDR were considered using an FDR < 0.05 as a threshold.

The protein–protein interaction results obtained after network diffusion for the top 100 proteins was visualised with the help of Cytoscape software (v 3.9.1)46.

Lastly, a kinase-substrate enrichment analysis (KSEA) was performed using the KEA3 web application47. The 100 proteins with the highest RWR score were used as input. Both rows and columns of the graph were clustered using rank by sum to obtain the top substrate and top kinase in the first row and column of the heatmap.

Immunofluorescence and cilia quantification

BJ-5ta cells: WT, BBS1 KO and ALMS1 KO were grown to confluence on coverslips, serum starved for 48h to promote primary cilia formation and fixed 10 min at RT in PBS + 4% PFA. To eliminate PFA residues, it was quenched with 100mM Glycine for 15 min at RT. Cells were then blocked and permeabilized for 60 min at RT in PBS + 0.1% Triton X100 + 3% BSA + 0.02% sodium azide (blocking solution). Coverslips were then incubated in a humidified chamber for 2 h at RT with blocking solution- diluted primary antibodies: anti-ARL13B (Proteintech, 17,711–1-AP, 1:1000,) and Anti-acetylated α-tubulin (AcTub) (Sigma, T7451, 1:5000). After three PBS washes, PBS- diluted secondary antibodies: anti-rabbit AF488 and anti-mouse AF594 (A-11008 and A-11005, Thermofisher, 1:1000) and DAPI (1 µg/ml, Thermofisher) were added for 1 h at RT in the dark. After three more PBS washes, coverslips were mounted on slides using Prolong Glass (Thermofisher), Cells were imaged using a Nikon NiE fluorescence microscope. Brightness and contrast of microscopic images were adjusted for optimal visualization with Fiji (Image J). To quantitate the percentage of ciliogenesis cells were counted in ten representative fields spanning the entire coverslip. At least 90 cells were counted per coverslip.

Results

ALMS1 and BBS1 depletion alters the phosphorylation of several proteins upon TGF-β1 ligand stimulation

To identify whether TGF-beta signalling was affected in the ALMS1 and BBS1 knockout cells and discover potentially new processes, we performed global phosphoproteomics in the presence and absence of TGF-β1 stimulation (methods). The experiment was designed using two multiplexes of 10 samples each. The batch 1 contains all samples without TGF-β1 stimulation, while batch 2 contains all samples stimulated for 30 min with TGF-β1 (Fig. 1A,B). We identified a total of 2,720 phosphopeptides (1,406 proteins) in batch 1 (samples without stimulation) and 3,317 phosphopeptides (1,619 proteins) in batch 2 (samples with stimulation), 1076 common proteins were identified in both batches (Fig. 1B). Of the total peptides/proteins identified, 50–60% were quantified in all samples of at least one condition (731 proteins in batch 1 and 960 in batch 2) (Figure S1 C-F).

Fig. 1
figure 1

Experimental design of the phosphoproteome in basal and TGF-β-stimulated BJ-5ta. (A) Protocol followed for the generation of the phosphoproteomic datasets (B) Overlap between the proteins identified in Batch 1 (Basal) and Batch 2 (TGF-β).

Full size image

After differential phosphorylation analysis, no significant proteins were detected in batch 1 for BBS1 knockout (BBS1KO) and 3 differentially phosphorylated proteins (DPP) were detected in the case of ALMS1 knockout (ALMS1KO) (Fig. 2A-C). Results from batch 2 showed 41 DPP in BBS1KO and 10 in ALMS1KO (Fig. 2B–D). Of the total DPPs detected in the cell models, 4 were common among them, 3 of them were infra-phosphorylated (EHBP1, CDC42EP3, RPL27A) while 1 (SLC38A1) was over-phosphorylated.

Fig. 2
figure 2

Differential phosphorylation analysis showed alterations in protein activation after TGF-β stimulation. Labelled proteins have an FDR < 0.01. (A) Volcano-plot of differentially phosphorylated proteins (DPP; FDR < 0.05) in BBS1 knockout before TGF-β stimulation (B) Volcano-plot of DPP in BBS1 knockout after TGF-β stimulation (C) Volcano-plot of DPP in ALMS1 knockout before TGF-β stimulation (D) Volcano-plot of DPP in ALMS1 knockout after TGF-β stimulation.

Full size image

Network diffusion highlights the involvement of ALMS1 and BBS1 in processes coordinated by the TGF-β pathway

Then, we performed a pathway enrichment analysis (methods) to identify the processes that are most affected by ALMS1 and BBS1 depletion after TGF-β stimulation (Fig. 3). Pre-diffusion ORA was performed using the DPP after TGF-β stimulation (only seed nodes). Due to the small number of proteins and the low signal-to-noise ratio of phosphoproteomics, we were unable to identify the TGF-β pathway, which serves as a control of the analysis (Fig. 3A,C). To improve the signal-to-noise ration from our DPP, we applied a network diffusion-based analysis (methods). The ORA after diffusion was performed with the top 100 proteins with the highest RWR score (Fig. 3B,D). In this case, we obtained TGF-β in both KO cell models. In addition, processes related to TGF-β regulation such as mitosis, NOTHC1, p53 or SHh signalling in the case of BBS1KO and endocytosis or Syndecan4 in the case of ALMS1KO were also identified (Fig. 3B,D).

Fig. 3
figure 3

Over-representation analysis (ORA) after network diffusion reveals processes controlled by TGF-β such as endocytosis, mitosis, or extracellular matrix in Bioplanet DB. (A) ORA of DPP in the BBS1KO before network diffusion. (B) ORA of top 100 proteins with the highest RWR score in the BBS1KO after network diffusion. (C) ORA of DPP in the ALMS1KO before network diffusion. (D) ORA of top 100 proteins with the highest RWR score in the ALMS1KO after network diffusion.

Full size image

Analysis of protein–protein interactions involves CDC42 and several RAB proteins in the alterations of TGF-β signalling in ALMS1 and CDK2 in BBS1

Using the top 100 proteins with the highest RWR score, an analysis of protein–protein interactions was performed. In the case of BBS1KO, CDK2 (seed node), a key regulator of the cell cycle, was one of the most interconnected nodes, highlighting its central role in this interactome (Fig. 4A). On the other hand, in the case of ALMS1KO, CDC42EP3 a direct effector of CDC42, key regulator of cell cycle and migration processes, was found to be one of the most interconnected nodes in the network, highlighting its central role in the deregulated interactome of this model. It also highlights the highly interconnected cluster of RAB proteins such as RAB11 and RAB8A, resulted by EHBP1, that play an important role in endocytosis (Fig. 4B).

Fig. 4
figure 4

Protein–protein interaction analysis highlights CDK2 as the central node of the BBS1 interactome and CDC42EP3 as the central node of the ALMS1 interactome. Red nodes represent the DPP that were detected in the first analysis. Blue nodes represent the nodes inferred by RWR algorithm. (A) Interactome of the top 100 proteins with the highest RWR score after network diffusion in the BBS1KO. (B) Interactome of the top 100 proteins with the highest RWR score after network diffusion in the ALMS1KO.

Full size image

Kinase-substrate enrichment analysis uncovers CDK2 and CDC42 as the most phosphorylated substrates

To determine which DPP were subject to the highest level of up-regulation, a KSEA was performed with the top 100 with the highest RWR score. The proteins (substrates) subject to the highest level of up-regulation were again CDK2 in the case of BBS1 and CDC42 in the case of ALMS1 (Fig. 5A,B). The main kinases regulating the selected subset of proteins were AKT1 in BBS1 and CDK2 in the case of ALMS1 (Fig. 5A,B). This shows that the observed alterations have commonalities but appear to differ in their regulation.

Fig. 5
figure 5

KSEA highlights CDK2 in BBS1 and CDC42 in ALMS1 as the main substrates subject to regulation. (A) KSEA of the top 100 proteins (substrates) with the highest RWR score after network diffusion in the BBS1KO. (B) KSEA of the top 100 proteins (substrates) with the highest RWR score after network diffusion in the ALMS1KO. The plots only show the 20 most up-regulated proteins (substrates) out of the initial 100 proteins used in the analysis.

Full size image

Discussion

The TGF-β pathway is one of the main signalling pathways regulated by the primary cilium2,48,49,50. It controls processes related to cell proliferation and migration and is associated with several pathologies such as liver or kidney fibrosis, recurrent phenotypes in ciliopathies32,51,52. Despite this, few studies have attempted to link ciliary gene dysfunction with signalling alterations in the TGF-β pathway33,34,35,53. In this study, we have applied phosphoproteomics to determine alterations in the TGF-β signalling pathway in two model ciliopathies, ALMS and BBS. We have further elucidated the regulation of the TGF-β pathway in ALMS and established for the first time the link between the TGF-β pathway and BBS genes33,34,35.

In this article, we designed and applied a pipeline that allowed us to obtain a global view of the processes altered in the TGF-β pathway by inhibiting two ciliary genes, ALMS1 and BBS1. For this purpose, we have used CRISPR-edited knockout models on a commercial line of immotarlised firboblasts (hTERT-BJ-5a). The hTERT-BJ-5ta cell line, derived from human neonatal fibroblasts and immortalised with hTERT, is a useful in vitro model to study fibroblast activation. Although these cells are not fibrotic, stimulation with TGF-β1 induces a profibrotic phenotype, including increased expression of α-SMA, collagen type I (COL1A1), and fibronectin. This response mimics key features of myofibroblast differentiation, allowing the study of fibrosis-related pathways in a controlled setting. Despite lacking the complexity of in vivo models, hTERT-BJ-5ta cells are widely used to investigate TGF-β1-driven fibrotic signalling54,55.

Regarding the ALMS1 KO, we found that the altered processes were mainly related to endocytosis and mitosis (via the Syndecan 4 signalling pathway, responsible for cytokinesis)56. The role of ALMS1 in endocytotic and mitotic processes has already been established in the past22,34,57,58,59. However, this is the first time that these alterations in endocytosis and mitosis could be related to the TGF-β signalling pathway affecting the activation of RAB11FIP5 and EHBP1 (seed nodes) and other related proteins such as RAB8A and RAB11B (inferred nodes) among others60,61. In addition, the relationship of CDC42 in this interactome has also been established through CDC42EP3, a CDC42 effector protein that is under-phosphorylated in the differentially phosphorylated proteins detected in the differential analysis (Fig. 2D, 4B). The significance of CDC42 has also been determined by KSEA (Fig. 5B). CDC42 plays a key role in cell polarity and regulation of mitosis62,63,64. CDC42EP3 acts downstream of CDC42 and regulates the rearrangement of the actin and septin cytoskeleton65. Cell cycle disorders and cytoskeleton alterations are common phenotypes in ALMS1-depleted cells, so a dysfunction of CDC42 and CDC42EP3 could be a possible cause of them34,58,59,66. Inhibition of CDC42EP3 and EHBP1 in BBS1KO could indicate commonalities in regulation between ALMS1 and BBS1, so that the alterations detailed above could also be present when BBS1 is inhibited.

BBS1 gene inhibition altered the phosphorylation of 41 proteins after TGF-β stimulation. Following network diffusion, these proteins were found to be mainly involved in the regulation of the extracellular matrix via the TGF-β pathway; cross-signalling with other pathways such as p53, NOTCH1 or SHh and transport of cation ions (Ca2+ or Na+) (Fig. 3B). Alterations of TGF-β signalling have already been described when inhibiting other ciliary genes such as ALMS1 or CEP128, but never for the BBS1 gene33,34,53. Some of the main nodes of the BBS1 interactome and KSEA were SYNPO2, CDK2 and PDGFRA proteins. SYNPO2, which is over-phosphorylated, encodes actin binding proteins that has been characterized as a tumor suppressor67. It is considered a protein with inhibitory capacity on cell migration and proliferation67. On the other hand, CDK2 and PDGFRA, which appear under-phosphorylated, are key regulators of mitosis and cell proliferation68,69,70. Thus, the lack of BBS1 seems to prevent proper TGF-B-mediated proliferation. Our data do not allow us to conclude the existence of a compensatory mechanism via another pathway.

Taken together, our results position CDC42 and CDK2 as convergent axes linking ciliary signalling with, respectively, vesicular trafficking and cell cycle checkpoints, thus providing a unified mechanistic framework for the alterations observed in ALMS and BBS. These alterations are congruent with the decrease in cilia formation that could be seen in both BBS1 and ALMS1 KO during phenotyping of the models (Figure S2).The main limitations of this study are related to the use of in vitro models, which often do not fully reproduce patient phenotypes. For this reason, these results need to be validated in primary samples or animal models to determine whether the alterations described are conserved processes.

On the other hand, phosphoproteomics is an approach that allows the global analysis of the activation of the cellular interactome. However, like any technique, it has several limitations, since the phosphoproteome is estimated at 30% of the total proteome and current enrichment techniques have a recovery of 1–3% of the total amount of starting sample71,72. The first limitation is due to a large amount of starting material needed to obtain the first data. The recommended starting material required is usually a minimum of 600µg-1mg of protein lysate73. This is affordable when working with cell culture, but very complicated when working with tissue samples. This makes it necessary to develop techniques that increase the enrichment yield and allow the use of less starting material. Phosphoproteomics data are usually scarce and noisy, requiring the development of statistical techniques and analysis methodologies that allow robust inference from the data initially obtained73. Furthermore, the nature of the data generated by current phosphoproteomics techniques introduces a bias in that kinases have only been identified from less than 5% of the phosphoproteome, and functional assignments of phosphosites are almost negligible74.

Conclusion

In conclusion, the depletion of ciliary genes such as ALMS1 and BBS1 alters signal transduction through the TGF-β pathway, altering processes such as endocytosis and mitosis in the case of ALMS1 and the regulation of the extracellular matrix, p53, NOTCH1, SHh and transport of cation ions in the case of BBS1. These phenomena could be explained by alterations affecting the inhibition of CDCD42 in the case of ALMS1 or CDK2 in the case of BBS1.