Dual proximity-based interactome mapping of FKBP51 and FKBP52 uncovers shared metabolic networks

Antibodies

The following antibodies were used: rat anti-FKBP51 (1:1000, R&D Systems MAB4094, RRID:AB_1964585), (1:1000 for western blot; 1:100 for staining, R&D Systems AF4094, RRID:AB_2103136), rabbit anti-FKBP51 (1:1000, Cell Signaling 12210S, RRID:AB_2797846), rabbit anti-FKBP52 (1:1000 for western blot; 1:500 for staining, Cell Signaling 11826S, RRID:AB_2797737), rat anti-Hsp90 (1:1000, Enzo ADI-SPA-835, RRID:AB_2039281), rabbit anti-V5 Tag (1:1000, Bethyl Laboratories, Inc A190120A RRID:AB_67586), mouse anti-Lamin (Proteintech 66095-1-Ig, RRID:AB_11232208), mouse anti-Alpha Tubulin (Proteintech 66031-1-Ig, RRID:AB_11042766), rabbit anti-PHB1 (Cell Signaling 2426S, RRID:AB_823689), rabbit anti-ALDH3A1 (Proteintech 15578-1-AP, RRID:AB_2273745), mouse anti-β Actin (Proteintech 66009-1-Ig, RRID:AB_2687938), and rabbit anti-β Actin (Proteintech 20536-1-AP, RRID:AB_10700003). HRP-tagged secondary antibodies for mouse (1030-05, RRID:AB_2619742), rabbit (4010-05, RRID:AB_2632593), and rat (3010-05, RRID:AB_2795801) were purchased from Southern Biotech (1:1000), except Streptavidin-HRP (1:4000, Thermo Scientific 21130). Highly cross-absorbed fluorescent secondaries were purchased from Invitrogen for donkey anti-goat Alexa Fluor 549 (A32758, RRID:AB_2762828) and donkey anti-rabbit Alexa Fluor 647 (A32795, RRID:AB_2762835) and used at 1:500. All primary and secondaries were prepared in 7% non-fat milk (Research Products International M17200) in Tris buffered saline (TBS) with 0.2% Triton X-100 (Research Products International 111036), except Streptavidin-HRP was prepared in TBS containing 3% bovine serum albumin (BSA) (Fisher BioReagents BP1600-100) and 0.2% Triton X-100. Mouse Anti-V5-tag Magnetic Beads (MBL International Corporation M167-11, RRID:AB_3716385) and Streptavidin Magnetic Beads (Thermo Scientific 88817) were used for co-immunoprecipitations.

Construction of the plasmids

V5-TurboID-NES_pCDNA3 was a gift from Alice Ting (Addgene plasmid 107169; http://n2t.net/addgene:107169; RRID:Addgene_107169). N- and C-terminus TurboID tagged FKBP51 (pCDNA3 V5-TurboID-GS-FKBP51-NES, pCDNA3 V5-FKBP51-GS-TurboID-NES) and C-terminus TurboID tagged FKBP52 (pCDNA3 V5-FKBP52-GS-TurboID-NES) plasmids were generated by GenScript through subcloning into the above plasmid with a 13 amino acid GS linker. The pCMV6 empty vector, pCMV6 FKBP51, and pCMV6 FKBP52 plasmids were previously generated in-house¹². Human, TOMM20 cDNA ORF Clone, C-GFPSpark^® tag was purchased from Sino Biological (HG15094-ACG).

TurboID construct testing

HeLa cells (ATCC CCL-2 RRID:CVCL_0030) were grown in MEM (Corning 10-010-CV, Corning, NY) plus 10% FBS (Sigma F4135-500ML), 1% penicillin/streptomycin (Cytiva SV30010), and 1% Gibco MEM non-essential amino acids (Gibco 11140050). For initial biotinylation evaluation and V5 co-immunoprecipitation, HeLa cells were plated at 300,000-400,000 cells/well into tissue-cultured treated 6-well dishes (Sigma SIAL0516) or at 2,000,000 cells/dish in 100 mm dishes (Starstedt 83.3902.300), transfected the following day with V5-TurboID-NES, V5-TurboID-GS-FKBP51-NES, and V5-FKBP51-GS-TurboID-NES and harvested 48 hours after transfection. For dexamethasone treatments, the cells were treated the same way but changed to charcoal stripped FBS (Sigma F6765) when plated and treated with either 0.5 μM dexamethasone (Sigma D2915) or vehicle (Dimethyl sulfoxide; DMSO, VWR 67-68-5) for the final 24 hours. For 6-well dishes, 2 μg of plasmid was transfected per well using 5 μL Lipofectamine 2000 (Invitrogen 11668019) and 15 μg of plasmid was transfected into 100 mm dishes with 30 μL of Lipofectamine.

For the biotinylation check, 50 μM biotin was added into the media for 30 minutes and 2 hours before harvest (Sigma B4501), followed by lysis in RIPA (Research Products International R26200) supplemented with Protease Inhibitor (Sigma P8340), Phosphatase Inhibitor Cocktail 2 (Sigma P5726), Phosphatase Inhibitor Cocktail 3 (Sigma P0044), and 1 mM PMSF (dissolved in 100% Ethanol, Sigma P7626). Cells were allowed to rest on ice for 20 minutes in lysis buffer and then centrifuged for 10 min at 13,000 ×g at 4 °C. Supernatant was transferred to a new tube and protein was measured by BCA (Pierce PI-23225, Rockford, Illinois). Western blots were run using 25 μg of each sample in 6x Laemmli SDS sample buffer (Thermo Scientific J61337AD) loaded into AnyKD gels (Bio-Rad 456903) and transferred onto Immobilon^™-P PVDF Membrane (Sigma IPVH00010), as previously described.⁴⁶

For V5 co-immunoprecipitation, the cells were harvested as described above but using M-PER (Pierce PI78505) supplemented with the protease and phosphatases inhibitors above. After BCA, 7 μg of total protein was saved for inputs and 500 μg of total protein from each sample was incubated with 50 μL of washed anti-V5-tag Magnetic Beads (MBL International Corporation M167-11, RRID:AB_3716385) for 2 hours on a nutator at 4 °C. Tubes were then placed on a magnet and magnetic beads were washed 3 times with M-PER buffer. Magnetic beads were then incubated with 30 μL of 2x Laemmli buffer (Bio-Rad 161-0737) with 5% βME (Thermo Scientific 125472500) and heated at 100 °C for 5 minutes. After returning tubes to the magnet, supernatant was collected and run on a western blot, as described above.

Samples for the streptavidin co-immunoprecipitation were prepared following the same experimental design including 30 minutes of 50 μM biotin exposure before harvest in RIPA buffer. After BCA, biotinylated proteins in 300 μg of cell lysate were affinity purified with 25 μL streptavidin magnetic beads for 1 hour. The protein bound beads were then thoroughly washed on nutating mixer (Corning 6720) with RIPA buffer for 2 minutes, 1 M KCl for 2 minutes, 0.1 M sodium carbonate for 10 seconds, 2 M urea in 10 mM Tris-HCl (pH 8.0) for 2 minutes, and 3 times with RIPA buffer for 2 minutes before boiling in 30 μL of 2x Laemmli buffer with 5% βME.

Subcellular fractionation

Subcellular fractionation was performed with the Mitochondria Isolation Kit for Cultured Cells (Abcam ab110171) according to the manufacturer directions with the following modifications. For each plasmid, four 150 mm dishes were plated at a cell density of 2.5 million cells per dish and transfected two days later with 30 μg of plasmid and 90 μg of Polyethylenimine (PEI) Max (Kyfora Bio 24765100). Forty-eight hours after transfection, cells were placed on ice, washed with 10 mL of ice-cold PBS, and then transferred using a cell scraper and 10 mL of ice-cold PBS into a pre-chilled 50 mL conical. This was repeated for each dish and combined into the same conical. The cells were then centrifuged at 500 ×g for 5 minutes at 4 °C to isolate the cell pellet. Cells were then frozen and thawed once on dry ice to weaken the cell membranes and resuspended with 1.6 mL of Reagent A (supplied in the kit) followed by a 10 min incubation on ice. Cells were then transferred to a pre-chilled 2 mL Dounce homogenizer, homogenized with 30 strokes, and transferred to a 2 mL centrifuge tube. Next the homogenate was centrifuged at 1,000 ×g for 10 min at 4 °C, and the first supernatant saved on ice. The pellet was then resuspended in 1.6 mL Reagent B (supplied in the kit), homogenized with 30 strokes in the Dounce homogenizer, and the homogenate centrifuged at 1,000 ×g for 10 min at 4 °C. The pellet (nuclear fraction) was resuspended in 550-800 μL of RIPA buffer supplemented with protease inhibitor. The supernatant was combined with the prior supernatant and centrifuged at 12,000 ×g for 15 min at 4 °C. After centrifugation, the supernatant (cytosolic fraction) was collected, and the pellet (mitochondrial fraction) resuspended in 500 μL of RIPA buffer supplemented with protease inhibitor. Total protein was measured by BCA and 20 μg of protein from each fraction was used for Western blot.

TurboID fluorescence localization

For fluorescence localization studies, HeLa cells were plated in 6-well plates. The next day, cells were transfected with C-GFPSpark TOMM20 and pCMV6 empty vector, FKBP51, or FKBP52 or pCDNA3 V5-TurboID-NES, V5-TurboID-GS-FKBP51-NES, or V5-FKBP51-GS-TurboID-NES using FuGENE HD (Promega E2311) according to the manufacturer’s instructions. After 1 day, cells were lifted with trypsin and seeded into Nunc Lab-Tek II 8-chambered slides (Thermo Scientific 155409). After 24 hours, cells were fixed with 4% paraformaldehyde (Sigma Aldrich 158127) in PBS for 10 minutes. Cells were permeabilized with −20 °C methanol (Fisher Scientific A433P-4) for 6 minutes and blocked with 5% donkey serum (Lampire Biological Laboratories 7332100, Pipersville, PA) with 0.1% Triton X-100 for 30 minutes. Cells were then incubated with primary antibodies (FKBP51 and FKBP52) and 5% donkey serum in PBS overnight. After 3, 5-minute PBS washes, cells were incubated with Alexa Fluor conjugated secondary antibodies listed above for 45 minutes, then washed twice in PBS for 5 minutes each. 4′,6-Diamidino-2-phenylindole dihydrochloride (DAPI; Sigma Aldrich 32670) was incubated for 2 minutes in PBS at a concentration of 2 μL/mL, followed by 3 5-minutes washes. After drying, slides were coverslipped using ProLong Diamond Antifade Mountant (Molecular Probes P36961). Fluorescent confocal images were generated using a Zeiss LMS880, AxioObserver laser scanning confocal microscope using a 63×/1.40 PLAN APO oil objective with a 1.6 optical zoom at a resolution of 8 bit (1192 x 1192 pixels) using 405 nm (9.40% power 405ex/462em), 488nm (9% power 488ex/554em), 594 nm (26.24% power 594ex/630em), and 633 nm (15.73% power 633ex/697em) with unidirectional scanning and a line average of 16. Images were then exported using Zeiss Zen 3.3 software.

TurboID proximity labeling and purification

For TurboID experiment used for mass spectrometry analysis, HeLa cells were plated at 5 million cells/dish in 150 mm dishes (Thermo Scientific 130183) for the TurboID labeling experiment for proteomic analysis. The day after plating, each dish was transfected with 45 μg of respective plasmid using 90 μL of Lipofectamine 2000. Directly before harvest, after a rinse with PBS, cells were exposed to 50 μM biotin for 30 minutes.⁴⁷ Cells were then washed five times with ice cold PBS and then collected into RIPA buffer with protease and phosphatase inhibitors. After 15 minutes on ice, lysates were centrifuged at 13,000 ×g for 10 minutes. Total protein was measured in the supernatant by BCA and 2,000 μg of protein from each sample was used for the immunoprecipitation with Streptavidin magnetic beads, which was performed in 1.5 mL low-protein binding tubes. Streptavidin beads were washed 3 times with RIPA buffer for 2 minutes (shaking) before being combined with the samples at 4 °C overnight. After a quick spin, the tubes were placed on a magnet and the supernatant removed. The beads were then thoroughly washed twice with RIPA buffer for 5 minutes (shaking), 1 M KCl for 5 minutes (shaking), 0.1 M sodium carbonate for 10 seconds, 2 M urea in 10mM Tris-HCl (pH 8.0) for 10 seconds, twice with RIPA buffer for 5 minutes (shaking), and twice with PBS for 2 minutes (shaking). After the last wash, beads were resuspended in 55 μL of 50 mM ammonium bicarbonate, 5% SDS, 50 mM dithiothreitol (DTT), 10 mM Biotin and then heated for 10 minutes at 95 °C in a ThermoMixer C (Eppendorf 5382000023, Enfield, CT) while shaking at 1000 rpm. Samples were then frozen at −80 °C until S-trap processing.

Label-free quantitative proteomics

Samples were processed for LC-MS/MS analysis using an S-Trap micro spin column (Protifi C002-MICRO-0010PK) according to the manufacturer’s protocol with minor adjustments as previously described.⁴⁷ In brief, eluted proteins were reduced with a final concentration of 70 mM DTT for 10 minutes, alkylated with 140 mM indole-3-acetic acid (IAA) for 30 minutes, and loaded onto an S-Trap micro spin column. Proteins were washed a total of 4 times with S-Trap binding/wash buffer containing 100mM Tris, columns were transferred to a new clean tube, then proteins were digested with 1 μg of Trypsin/Lys-C (Promega). Samples were eluted into one pool first by 50 mM ammonium bicarbonate, then by 0.2% formic acid in LC-MS grade water, and finally by 50% acetonitrile (ACN) containing 0.2% formic acid. Samples were diluted in 0.1% formic acid in water to reduce ACN concentration for subsequent desalting by C18 SPE columns (Waters). Once loaded, samples were washed with 0.1% formic acid in water and eluted in 50% acetonitrile containing 0.1% formic acid. Samples were then dried down with a speed vacuum manifold and cold vapor trap and, finally, resuspended in 0.1% formic acid in water for LC-MS/MS analysis.

Samples were analyzed by diaPASEF on a timsTOF Pro instrument coupled to a nanoElute UHPLC system (Bruker). A CaptiveSpray ion source with column oven heated to 50 °C containing an Aurora Ultimate CSI UHPLC reversed-phase C18 column (25 cm × 75 μm i.d., 1.7 μm C18, IonOpticks AUR3-25075C18-CSI) was utilized. Mobile phases A (0.1% formic acid in water) and B (0.1% formic acid in acetonitrile) were used in a 45-min gradient of 2-25% B (total run time of 60 min including a 25-37-90% B step ramp up and hold at 90% B for 10 min to fully clean the column prior to the next sample). The diaPASEF scan mode was implemented within an ion mobility range of 0.7- 1.40 1/K₀ [V·s/cm²] spanning 250-1425 m/z with a 1.48 s estimated cycle time. Ion mobility and m/z calibrations were performed linearly using 3 ions at 622 m/z, 922 m/z, and 1222 m/z (Agilent).

Data-independent acquisition (DIA) data were analyzed with DIA-NN (version 1.8.1)³⁷ using a predicted library generated from the UniProt Homo sapiens database (78120 entries). Label-free quantification (LFQ) was applied with match-between-runs (MBR) using a 1% FDR (global precursor and protein) cutoff. Additionally, 15 ppm mass/MS1 accuracy, genes protein inference, single pass mode, robust LC (high precision) quantification strategy, RT-dependent cross-run normalization, and smart profiling options were applied.

Proteomic statistical filtering

Differential abundance analysis was conducted in Mass Spectrometry Downstream Analysis Pipeline (MS-DAP, v. 1.2.2) within RStudio where DIA-NN output (full report) and the associated FASTA file were input along with specific sample metadata.³⁸ Three contrasts (FKBP51-TurboID/TurboID Control, FKBP52-TurboID/TurboID Control, and FKBP52-TurboID/FKBP51-TurboID) were specified for differential abundance analysis using the algorithm MSqRob.⁴⁸ Minimum detection and quantification filters of at least 4 out of 5 replicates per group were utilized along with variance Stabilizing Normalization (VSN) and then Mode Between protein (MBprot) normalization steps prior to differential abundance analysis. The MSqRob q-value and log2 foldchange threshold was set to less than 0.01 and to automatically infer through bootstrapping, respectively. These filtering criteria were used to achieve a statistically robust list of differentially abundant proteins (DAPs labeled as “TRUE” in output). Given that imputation to replace missing values was not performed in the DAA workflow, uniquely represented proteins in the FKBP51-TurboID and FKBP52-TurboID groups in which detection and quantification did not meet the specified filters above were reported based on a differential z-score for observed peptides. Proteins that had a q < 0.01 and Log2 Fold Change > 0.58 (which exceeded the bootstrapping-determined fold change cutoff in MS-DAP) or z-score > 6 were used for bioinformatic analysis. The z-score heatmap with dendrogram was generated in R v4.5.2 using the ComplexHeatmap package (v2.26.0),^44,45 with color mapping performed using circlize package (v0.4.17). Row clustering was conducted using hierarchical clustering in base R (stats::hclust) with complete linkage on Euclidean distances (stats::dist), computed using the FKBP52-TurboID column only (columns were not clustered).

Bioinformatic analysis

STRING analysis⁴⁹ was performed to generate Cellular Component analysis using the functional enrichment visualization tool with the enriched gene list for each data set. Data were sorted using default setting by −log(FDR), except rows were merged by term similarity ≥90%. STRING 12.0 visualization⁴⁹ was performed in Cytoscape⁴¹ Version 3.10.4 using the StringApp⁴² (physical subnetwork, Confidence 0.4) using the list of enriched proteins from the proteomic analysis in this study or interaction lists for FKBP51 and FKBP52 from BioGrid⁵⁰ that have physical evidence from at least 1 study. Node connections represent interactions with physical evidence using the genome as the background for functional enrichment. Nodes were then sorted using the Prefuse Force Directed Layout with default settings into subgroups and then further organized by a combination of circular, hierarchical or stack node layouts to improve visualization. Vector graphics were then exported. Circos plots⁵¹ were generated in Metascape⁴⁰ using the same lists as above from this study and BioGrid using default parameters. Protein-protein interaction enrichment analysis was performed on the same lists in Metascape with default parameters (Min Overlap: ≥ 3, P Value Cutoff: ≤ 0.01, Min Enrichment: 1.5), which uses multiple protein interaction databases (STRING, BioGrid, OmniPath, InWeb_IM)^39,50,52 and then the Molecular Complex Detection (MCODE) algorithm⁵³ was applied. KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis was performed using TurboID enriched proteins using ShinyGO 0.85 with the default settings (FDR ≤ 0.05, Pathway size: Min 2).⁴³ Enrichment heatmaps were generated using default parameters in Metascape v3.5.20260201 using the TurboID enrichment datasets.

Mitochondrial stress test

For mitochondrial stress assays, HeLa cells were cultured as described above. cells were plated at 300,000 cells/well in 6-well plates and, 24 h later, transfected with 2 μg of V5-TurboID-NES, V5-FKBP51-GS-TurboID-NES, V5-FKBP52-GS-TurboID-NES, pCMV6-FKBP51, pCMV6 FKBP52, or pCMV6 empty vector using FuGENE HD (Promega E2311), following the manufacturer’s instructions. After 48 h, transfected cells were harvested and reseeded at 10,000 cells/well onto 0.01% poly-L-lysine–coated Seahorse XF96 microplates (6–8 technical replicates per condition; Agilent Technologies 103793-100) and incubated overnight at 37 °C. Sensor cartridges were hydrated and the XF96 analyzer was warmed up overnight according to the manufacturer’s recommendations.

On the assay day, culture medium was replaced with pre-warmed Seahorse XF base medium (Agilent Technologies 103575-100) supplemented with 10 mM glucose (Agilent Technologies 103577-100), 1 mM pyruvate (Agilent Technologies 103578 −100), and 2 mM L-glutamine (Agilent Technologies 103579-100), followed by 1 h equilibration in a 37 °C non-CO₂ incubator. Cartridges were loaded with Mito Stress Test inhibitors (Agilent 103015-100): port A, oligomycin (15 μM stock; 1.5 μM final); port B, FCCP (10 μM stock; 1 μM final); port C, rotenone/antimycin A (5 μM stock; 0.5 μM final). Oxygen consumption rate (OCR) was measured on an Agilent Seahorse XF96 analyzer over sequential basal (3 cycles), oligomycin (3 cycles), FCCP (3 cycles), and rotenone/antimycin A (3 cycles) injections (3 min measure, 3 min mix per cycle).

Data were analyzed in web-based Wave software (Agilent), excluding outliers based on instrument quality control flags and values outside mean ± 3 SD within groups. After the assay, plates were briefly centrifuged at 300 ×g, washed with PBS, and lysed in ice-cold RIPA buffer containing protease and phosphatase inhibitors, then stored at −20 °C overnight. Total protein was quantified using the Pierce BCA Protein Assay kit (Thermo Scientific 23225) and used to normalize OCR parameters (pmol/min/μg protein). Replicates in each experiment were averaged into a single representative value, which was analyzed with values from four independent biological replicates.

Visual representation and statistical analysis

Schematics were created using BioRender.com. Western blot band intensities were quantified with BioRad Image Lab software. GraphPad Prism v11.0.0 (GraphPad Software, San Diego, CA, USA) was used for generation of volcano plots and bar graphs. Significant differences were conducted in GraphPad Prism using a one-way analysis of variance (ANOVA) followed by Dunnett's post hoc test as indicated in the figure legends. p-values less than 0.05, 0.01, 0.001 or 0.0001 are marked by 1, 2, 3, or 4 asterisks (*), respectively. Error bars represent standard error of the mean (SEM).

TurboID tagging of FKBP51 and FKBP52 promotes biotin labeling without disrupting FKBP-Hsp90 interaction

To determine the optimal location of adding a TurboID tag onto FKBP51, plasmids were generated with either an N- or C-terminal TurboID, connected by a GS linker, which allows biotin labeling of proximal proteins (Figure 1A). A nuclear export signal (NES) was included on the C-terminus, which supports labeling in a wide variety of cellular compartments.^54,55 Co-immunoprecipitation of the V5 tag from HeLa cells transfected with FKBP51-TurboID or TurboID-FKBP51 confirmed that the TurboID tag did not block complex formation with Hsp90 (Figure 1B). Both constructs show robust expression of FKBP51 conjugated to TurboID at ~90kDa, which corresponds to the expected molecular weight of FKBP51 attached to TurboID. As a promiscuous biotin-ligase, TurboID, will label proteins within a ~10-20 nm proximity,^35,56 even if only a transient interaction occurs. Proteins outside of this TurboID biotinylation zone will not be biotinylated (Figure 1C). Both N- and C-terminal TurboID fused to FKBP51 resulted in enhanced biotinylation of many proteins within 30 minutes of biotin addition (Figure 1D). Increased labeling time beyond did not result in additional biotinylation by western blot, so the lowest labeling time of 30 minutes was used in subsequent experiments. Additionally, biotin labeled proteins were efficiently enriched by streptavidin magnetic beads validating the technical workflow (Supplemental Figure 1). Based on these results, C-terminal TurboID constructs were selected for use in further experiments, in order to minimize disruption of PPIase activity at the N-terminus while maintaining C-terminal binding to known interactors.⁵⁷

To enable the identification of proximal proteins that are unique to FKBP51 and those overlapping with homolog, FKBP52, C-terminal TurboID constructs were made for both proteins (Figure 1E). Next the expression of TurboID, FKBP51-TurboID, and FKBP52-TurboID was compared to untagged FKBP51 and FKBP52 plasmids driven by a CMV-promoter, along with an empty vector control in the presence and absence of dexamethasone. For FKBP51, the expression level of FKBP51-TurboID was comparable to the dexamethasone-induced level, unlike the expression of untagged FKBP51, which showed robust overexpression ~35-fold higher than the endogenous level (Figure 1F, Supplemental Figure 2). For FKBP52, the expression level of FKBP52-TurboID was almost 4-fold higher than endogenous, but well below the ~13-fold enhancement observed for untagged FKBP52. Based on V5-tag quantification, expression levels of FKBP52-TurboID were significantly higher than TurboID alone, but not significantly altered compared to FKBP51-TurboID. To ensure the TurboID tag does not interfere with FKBP52 interaction with known partner Hsp90, co-immunoprecipitation of the V5 tag from transfected HeLa cells was repeated with the addition of FKBP52-TurboID, confirming the tag did not interfere with the complex (Figure 1G).

TurboID attachment to FKBP51 and FKBP52 preserves subcellular location

Given overexpression and fusion proteins can lead to altered cell location, subcellular fractionation and immunofluorescence staining was used to verify protein localization after transient transfection of each plasmid. The primary location for endogenous FKBP51 and FKBP52 in HeLa cells was cytosolic, with approximately 5-10% of the protein located in the nucleus and mitochondria relative to the cytosolic fraction, which was unaltered upon TurboID expression (Figure 2A-B). Only transfection of untagged FKBP51 and FKBP52 led to subcellular location changes, with FKBP51 demonstrating increased mitochondrial presence and FKBP52 showing increased mitochondrial and nuclear occupancy, which is likely more a consequence of abundance than a shift in location. Immunofluorescent staining in HeLa cells also confirmed that the TurboID tag did not interfere with protein subcellular localization of FKBP51 or FKBP52 compared to overexpression of untagged constructs (Figure 2C). Both FKBP51 and FKBP52 show broad cytosolic staining, including some appearance of overlap with mitochondrial marker, Tomm20. FKBP51 appears to also demonstrate nuclear localization, however, a portion of this signal may be non-specific, since it is also found in no primary controls to a lesser degree (Supplemental Figure 3). Similar to western blots the untagged FKBP51 and FKBP52 constructs show more robust expression than those with the TurboID tags.

FKBP51 shows unique and overlapping proximal proteins with homolog, FKBP52

After 48 hours of expression in HeLa cells and 30 minutes of biotin supplementation, cells expressing TurboID, FKBP51-TurboID, and FKBP52-TurboID were collected, affinity purified, and processed by mass spectrometry analysis. A total of ~3500 unique proteins were detected in each sample (Supplemental Figure 4). A similar or greater number of proteins was identified in the TurboID control compared to FKBP51-TurboID or FKBP52-TurboID samples, which enables high confidence in identified interactors.⁵⁸ A volcano plot was used to visualize the most enriched proteins from the FKBP51-TurboID (Figure 3A) and FKBP52-TurboID (Figure 3B) datasets. Proteins were considered significantly enriched if they were identified in at least 4 of 5 samples in each group and had a q-value < 0.01 and a Log2 Fold Change < 0.58, which represents a ≥1.5-fold enrichment. In the volcano plot and subsequent sections, gene names are used since they are standardized for the proteins they represent and are used consistently for bioinformatic analysis. As expected, many of the most enriched proteins have been previously identified, including the glucocorticoid receptor, NR3C1. As statistical comparison of q-value and Log2 Fold Change requires detection in both the experimental and control conditions, some proteins were not detected in the TurboID samples but were significantly enriched by FKBP51-TurboID and/or FKBP52-TurboID, as demonstrated by a z-score > 6 (Figure 3C). This suggests specific enrichment of these proximal proteins driven by the presence of FKBP, so they were included for subsequent analysis. This included NUDCD3, PINK1, SIK2, KYAT3, SETMAR, and DNA2, which showed a z-scores greater than 6 for both FKBP51-TurboID and FKBP52-TurboID. A number of other proteins were uniquely enriched by FKBP52-TurboID, including LRRK1, LIMK2, STK33, SNAPC4, C5ORF34, and NOD1. After filtering, 109 proteins were identified to be enriched in the FKBP51-TurboID samples, and 236 proteins were enriched in the FKBP52-TurboID samples compared to the TurboID control (Figure 3D; Supplemental Tables 1-4). Seventy proteins were identical between FKBP51-TurboID and FKBP52-TurboID. Assessment of the protein-protein interaction network in Metascape using the MCODE algorithm revealed multiple pathways that have been previously linked to FKBP51 and FKBP52, including the most prominent enrichment node corresponding to isomerase activity and telomere maintenance (Figure 3E), and additional nodes related to RNA surveillance and spliceosome regulation, Hsp90 binding and protein folding, and RNA polymerase activity (Supplemental Figure 5).

Filtered enrichment analysis delineates known and novel FKBP51 and FKBP52 interactors

To evaluate the validity of proximal proteins and compare previously reported protein interactors, a Circos graph was used to show the overlap of FKBP51-TurboID and FKBP52-TurboID enriched proteins with those reported on BioGrid. This analysis revealed that in addition to the 70 proteins identical between the FKBP-TurboID datasets, there were 31 proteins enriched in FKBP51-TurboID and 20 proteins enriched in FKBP52-TurboID that have been previously reported on BioGrid (Figure 4A; Supplemental Table 5). Interestingly, 8 proteins enriched by FKBP51-TurboID were previously identified for FKBP52 on BioGrid, while 58 proteins enriched by FKBP52-TurboID were previously linked to FKBP51 on BioGrid. There were 5 proteins that were enriched by both FKBP51-TurboID and FKBP52-TurboID that have also been previously confirmed by other studies to interact with both FKBP51 and FKBP52 including the glucocorticoid receptor NR3C1, the Hsp90 co-chaperones, PTGES3 and CDC37, the mitochondrial PTEN-induced kinase 1, PINK1, and the calcyclin binding protein, CACYBP. In addition to these identical protein matches, numerous proteins shared gene ontology terms across datasets, indicating substantial overlap. There were 78 interactors for FKBP51-TurboID and 216 interactors for FKBP52-TurboID not previously identified by physical evidence on BioGrid (Supplemental Table 5), demonstrating the effectiveness of proximity-based identification in expanding our understanding of protein-protein interactions. Overall, the identification of validated FKBP51 and FKBP52 interactors supports the rigor of this study and indicates that many novel interactors discovered here are likely meaningful.

In addition to specific protein partners, assessment of the protein-protein interaction network in Metascape using the MCODE algorithm revealed multiple pathways that have been previously linked to FKBP51 and FKBP52 (Figure 4B-D). Identified pathways include proteins associated with heat shock protein binding and protein kinase and phosphotransferase activity (Figure 4B), DNA repair and metabolism (Figure 4C), and estrogen response element binding and cellular response to steroid hormone (Figure 4D). Some highlighted enriched proteins across analyses include the TPR containing proteins: PPID, TTC1, TTC4, and RPAP3; the DNA repair machinery: MCM2, MCM5 and MCMBP; and the CHORD-Sgt1 (CS) domain containing proteins: NUDC, NUDCD3, NUDCD2, SUGT1, and CACYBP, which have been shown to interact with Hsp90.⁵⁹ Interestingly, USP19 is another CS domain containing protein that was specifically enriched by FKBP52-TurboID, replicating prior FKBP52-specific findings.⁶⁰ STRING analysis based on physical evidence reveals interconnected networks that are formed by the identified proteins for both FKBP51-TurboID (Supplemental Figure 6) and FKBP52-TurboID (Supplemental Figure 7). At the same time, there are many enriched proteins that have no known physical networks, which present opportunities for future research into their relationships.

FKBP51 and FKBP52 interact with proteins that regulate multiple aspects of metabolism

Further analysis using the STRING functional visualization tool to analyze the Gene Ontology (GO) Cellular Component of the FKBP51-TurboID and FKBP52-TurboID dataset revealed high enrichment of proteins associated with the cytosol and cytoplasm (Figure 5A-B), in agreement with the subcellular location analysis. In addition to the primarily cytosolic interactors, FKBP51-TurboID showed an enrichment of proteins associated with extracellular exosomes (Figure 5A), whereas FKBP52-TurboID showed enrichment associated with nuclear lumen and nucleus (Figure 5B). Analysis of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using ShinyGO reveals significant enrichment in metabolic pathways for both FKBP51-TurboID and FKBP52-TurboID (Figure 5C-D). Metascape analysis of enriched terms (GO/KEGG) further expanded on the findings above, revealing significant enrichment in both FKBP51-TurboID and FKBP52-TurboID for heat shock protein binding and metabolic processes including biosynthesis of amino acids, DNA metabolic process, and positive regulation of catabolic process (Figure 5E). FKBP52-TurboID samples were uniquely enriched in proteins associated with protein localization to organelle, primary neural tube formation, RNA polymerase complex, DNA-templated transcription, and nucleobase-containing compound catabolic process.

Mitochondrial function is preserved upon FKBP51 and FKBP52 overexpression, with confirmed ALDH3A1 co-immunoprecipitation

Since the most significant enrichment term for both FKBP51-TurboID and FKBP52-TurboID datasets was metabolism, the Seahorse XF analyzer was used to assess if this was related to functional changes in mitochondrial respiration. No significant alterations in HeLa cell respiration was measured by FKBP51 or FKBP52 overexpression, regardless of the presence or absence of the TurboID tag (Figure 6A-C). Despite the lack of functional change, co-immunoprecipitation of the V5 tag from HeLa cells transfected with FKBP51-TurboID or FKBP52-TurboID confirmed the association of FKBP51 and FKBP52 in a complex with mitochondrial regulator, ALDH3A1 (Figure 6D).

Limitations of the Study

Although this study provides significant insight into FKBP51 and FKBP52 interactors, there are limitations, especially in the validations of hits, expression choice, and cell model. Robust filtering criteria were used to identify potential interactors, but further validation is needed to confirm protein interactors. This is not trivial, since TurboID is sensitive enough to capture transient interactions, which will not be measurable by traditional protein complex measurements, like co-immunoprecipitation. It also should be noted that using overexpression and fusion proteins can lead to non-physiological protein interactions and cellular mislocalization, but FKBP51-TurboID and FKBP52-TurboID showed similar cellular compartment patterns. Additionally, given increased FKBP51 levels are often associated with age- and stress-related disorders,^32,84 dexamethasone-induced FKBP51 expression was used as a benchmark. Consistent with the subcellular localization, FKBP51-TurboID did not show significantly higher expression than dexamethasone-induced FKBP51. While the current transfection protocol used is fast and efficient, it prevents FKBP51 expression from being regulated by native stress-inducible promotors. Future studies would benefit from developing knock-in or endogenous tagging to evaluate stress-induced protein complexes.⁸⁵

Questions remain as to whether the protein interactors identified in this study are universal across other cell types. The use of HeLa cells was chosen because of their robust GR response.³⁶ However, cancer cells have enhanced metabolic activity including a preference for glucose metabolism.⁸⁶ High glycolytic activity typically favors mTOR activation, which when chronically upregulated can lead to metabolic dysfunction.^87,88 This enhanced glycolysis could influence the proportion of metabolic proximal proteins identified, while also explaining why mitochondrial function was unaffected by FKBP51 or FKBP52 overexpression. Future studies would benefit from studying non-cancerous cell lines, non-transformed cell lines, and extending to mouse models to investigate cell-type and tissue specificity. Further studies in mouse models may also aid in elucidating complex stress-responsive protein interactome shifts, as many age- and stress-related disorders are multifactorial.⁸⁹ Lastly, many of the protein interactors identified in this study are likely part of a multimeric complex with Hsp90. Future TurboID studies could shift the interactome to identify Hsp90-independent interactors by using mutations or isoforms that lack Hsp90 binding.⁴