Proximity Labeling of D1-Like Dopamine Receptors Captures Distinct Cellular Environments and Uncovers Trafficking Proteins That Regulate DA-Mediated Behaviors in Drosophila

Cloning

The coding regions of Drosophila Dop1R1 (isoform D) and Dop1R2 (isoform A) were PCR amplified from pUAST-Dop1R1 (gift from Josh Dubnau) and pUAST-Dop1R2 (called pUAST-DAMB) transgenic lines, respectively, using primers with FseI restriction sites in their tail. The resulting products were ligated into a singular FseI site in pCAGWBA, a pcDNA3.1-based expression vector to create pCAG-Dop1R1 and pCAG-Dop1R2 (untagged). The “empty” control was the pCAGWBA vector with nothing inserted into the FseI site. To generate SNAP and Turbo-V5 receptor fusions, we used Infusion cloning (Takara Bio). The SNAP tag and TurboID coding sequences were PCR amplified from pSNAPf Vector (New England Biolabs, N9183S) and pTurboID-His6_pET21a plasmid (Addgene, 107177), respectively, using primers with added tails containing a flexible linker (GGSGGSGGSGGS), or not, and containing 15 bp homology to intended C or N termini of the receptors or to the pCAG vector flanking the FseI site. Tag and receptor PCR products were combined with FseI linearized pCAG and ligated together, transformed into DH5a bacterial cells prior to sequencing verification of single clones. PM-Turbo-V5 was cloned similarly, but we inserted the Lyn11 peptide (MGCIKSKGKDSA) upstream of Turbo-V5 using primers and Infusion cloning.

Transfection

We cultured and transfected HEK293T/17 cells because of high efficiency for transfection and prior characterization of Drosophila dopamine receptor signaling in human contexts (see Supporting Information for details). Plasmids were transiently transfected using PLUS (10 μL/6 cm dish or 62.5 μL/15 cm dish) and Lipofectamine LTX (12 μL/6 cm dish or 75 μL/15 cm dish) following ThermoFisher protocol. Transfection media was added to cells for ∼21 h prior to characterizing expression or biotin labeling experiments. In the beginning (Figures A–C and S1A–E) we used equivalent DNA mass (2.52 μg/6 cm dish or 8.94 μg/15 cm dish) for empty, receptor, receptor fusion, or PM-Turbo-V5 plasmids but found that R2-Turbo-V5 had reduced expression. For Figure S1F, as well as proteomic experiments Figure , we adjusted DNA mass for only R2-Turbo-V5 to be 15.75 μg/15 cm dish (PM:R1:R2, 1:1:1.76). For live-cell imaging, empty vector and dopamine receptor constructs were co-transfected with PTX-S1 as done previously, alongside either pGloSensor-22F (for cAMP, Promega) or CalfluxVTN (for Ca²⁺), at a ratio of 6:1:6 or 6:1:1, respectively.

Live Cell Imaging for cAMP and Ca²⁺

For both cAMP and Ca²⁺ signaling assays, cells were used ∼21 h post transfection, and reporters were imaged with CLARIOstar microplate reader (BMG Labtech) at room temperature. Details for cell processing and imaging procedures can be found in Supporting Information.

Cell Staining and Microscopy

For Turbo fusion and biotinylation staining (Figure S1C), 300,000 cells in 500 μL were seeded onto poly-l-lysine coverslips placed in 12 well plates and grown until ∼50–60% confluency and then transiently transfected with 1 μg/well of pCAG vectors containing Dop1R1 and Dop1R2 with or without GGS₄-Turbo-V5 fused to the tail, as well as PM-Turbo-V5 and empty vector control. After ∼19 h post transfection, cells were either incubated in 500 μM biotin (using D-Biotin, Life Technologies, B20656) in normal culture media or not (normal culture media alone) for 1.5 h at 37 °C, 5% CO₂. Biotin media was removed and cells were fixed and stained. See Supporting Information for details on staining and microscopy for immuno-, biotin, and SNAP tag staining.

Preparing Turbo-Labeled Lysates for Western Blot and Proteomics

For generating Turbo-labeled lysates for Western blot (WB) and proteomic analysis, cells were first cultured and transfected as described above using 1 × 6 cm (cell #) or 4 × 15 cm (cell #/dish) dishes, respectively. To measure labeling across biotin media exposure time, cells were given biotin media (500 μM, using D-Biotin, Life Technologies, B20656) for 0, 3, 10, or 40 min prior to lysation (Figure S1D) and ran in three independent experiments (quantitated in Figure S1E); 40 min exposure was chosen for the remaining WB (Figure S1F) and MS experiments (Figures and S2). For ligand activation specific labeling (Figures and S2), cells are exposed to biotin 10 min (of the 40) prior to DA exposure (via media change) to increase biotin concentrations around the Turbo environment at the time of activation. Therefore, total ligand exposure was 30 min prior to stopping the reaction and lysing the cells. After biotin and/or DA treatment, cells were processed to make lysates for bead incubation or Western blot analysis (see Supporting Information).

Streptavidin Bead Purification

Bead purification based on published Turbo-related methods. Briefly, streptavidin magnetic beads (Dynabeads MyOne Streptavidin T1, ThermoFisher, #65601) were pre-washed as recommended by the manufacturer using the above lysis buffer and then incubated with lysates in a microfuge tube (1 μL beads/4.3 μg protein) on a nutator on random agitation with end-over-end rotation for 16–24 h at 4 °C. Beads were then washed and separated at 4 °C in the following solution/sequence: lysis buffer (×2), 1M KCl (×1), 100 mM Na₂CO₃ (×1), 2M Urea in 10 mM Tris-HCl pH 8 (×1), and lysis buffer (×2). Proteins were then eluted from the beads by resuspension in elution buffer (0.1% SDS, 2 mM biotin), and then the beads were vortexed for 10 s, boiled at 95 °C for 3 min, repeated 3 times. After separation, the eluants contained ∼20–40 μg of purified biotinylated proteome per sample measured by BCA assay.

LC–MS/MS Protein Identification and Quantification

For the TMT data set, eluted proteins (6 experimental conditions, 5 biological replicates/condition) were precipitated overnight with 4× ice-cold acetone (v:v). Protein pellets were then processed for trypsin digestion using micro S-Traps (Protifi, Huntington, NY) according to the manufacturer’s instructions. Briefly, the dried protein was resuspended in 22 μL of 5% SDS, reduced using DTT at 55 °C for 10 min, alkylated using methyl methanethiosulfonate (MMTS) at room temperature for 10 min, and then spun down at 20K for 10 min. Subsequently, 2.5 μL of phosphoric acid was added to the sample, followed by 165 μL of a mixture of HPLC-grade methanol and ProtiFi binding/wash buffer. Following loading onto the S-Trap, samples were washed three times using centrifugation and trypsin was added in 50 mM TEAB at a 1:25 w/w ratio. The S-Trap column was incubated for 1 h at 47 °C. Following this incubation, 40 μL of 50 mM triethylammonium bicarbonate (TEAB) was added to the S-Trap and the peptides were eluted using centrifugation. Elution was repeated once. A third elution using 35 μL of 50% acetonitrile (ACN) was also performed, and the eluted peptides were dried under vacuum. Peptides were reconstituted in 50 mM TEAB, and their concentrations were determined using the Pierce quantitative fluorometric peptide assay (Thermo Fisher Scientific, Waltham, MA). Six micrograms of peptides were labeled with TMT labels (6-plex) according to the manufacturer’s instructions (Thermo Fisher Scientific, Waltham, MA) and pooled. The pooled plexed samples were dried under vacuum, resolubilized in 1% TFA, and desalted using OASIS HLB 1cc solid phase extraction cartridges (Waters, Milford, MA), and then dried once again using vacuum. The dried labeled peptide plexed samples were then resuspended in 400 μL of 1M TEAB and fractionated using an Agilent 1100 HPLC system and high pH reversed phase chromatography on a Zorbax Eclipse XDB-C18 column (4.6 × 150 mm, 5-micron) from Agilent (Santa Clara, CA). Sixty fractions were collected at 0.5 mL/min over 105 min using a gradient of 0–5% solvent B in 10 min, 5–35% solvent B in 60 min, 35–70% solvent B in 15 min, a 10 min hold of 70% solvent B, and finally a return to 5% solvent B in 10 min. Solvent A consisted of 100 mM TEAB and solvent B consisted of ACN. The 60 fractions were dried vacuum, concatenated into 6 fractions, and then cleaned up using 2 μg capacity C18 ZipTips (Millipore, Billerica, MA) according to the manufacturer’s instructions and dried.

Dried TMT-labeled peptides were reconstituted in 5 μL of 0.1% formic acid and on-line eluted into a Fusion Tribrid mass spectrometer (Thermo Scientific, San Jose, CA) from an EASY PepMap RSLC C18 column (2 μm, 100 Å, 75 μm × 50 cm, Thermo Scientific, San Jose, CA) using a gradient of 5–25% solvent B (80/20 acetonitrile/water, 0.1% formic acid) in 180 min, followed by 25–44% solvent B in 60 min, 44–80% solvent B in 0.1 min, a 5 min hold of 80% solvent B, a return to 5% solvent B in 0.1 min, and finally a 20 min hold of solvent B. All flow rates were 250 nL/min delivered using an nEasy-LC1000 nano liquid chromatography system (Thermo Scientific, San Jose, CA). Solvent A consisted of water and 0.1% formic acid. Ions were created at 1.7–2.4 kV using an EASY Spray source (Thermo Scientific, San Jose, CA) held at 50 °C. A synchronous precursor selection (SPS)-MS3 mass spectrometry method was used by scanning between 380 and 2000 m/z at a resolution of 120,000 for MS1 in the Orbitrap mass analyzer and performing CID at top speed in the linear ion trap of peptide monoisotopic ions with charge 2–8, using a quadrupole isolation of 0.7 m/z and a CID energy of 35%. The top 10 MS2 ions in the ion trap between 400 and 1200 m/z were then chosen for HCD at 65% energy and detection in the Orbitrap at a resolution of 60,000 and an AGC target of 1E5 and an injection time of 120 ms (MS3).

Quantitative analysis of the TMT experiments was performed simultaneously with protein identification using Proteome Discoverer 2.4 software. The precursor and fragment ion mass tolerances were set to 10 ppm and 0.6 Da, respectively. Enzyme cleavage was performed by trypsin with a maximum of 2 missed cleavages, and UniProt Human proteome FASTA file (downloaded in 2017) with six additions (P00761|TRYP_PIG, P22629|SAV_STRAV, P02769|ALBU_BOVIN, P06709|BIRA_ECOLI, Q24563|DOPR2_DROM, P41596|DOPR1_DROME) was used in SEQUEST searches. The impurity correction factors obtained from Thermo Fisher Scientific for each kit were included in the search and quantification. The following settings were used to search the data: dynamic modifications: oxidation / +15.995 Da (M), deamidated / +0.984 Da (N, Q), biotin / +226.078 Da (K), TMT6plex / +229.163 Da (K) and static modifications of TMT6plex / +229.163 Da (N-terminus), carbamidomethyl +57.021 (C). Only unique+Razor peptides were considered for quantification purposes. The Percolator feature of Proteome Discoverer 2.4 was used to set a false discovery rate (FDR) of 0.01. The co-isolation threshold and SPS mass matches threshold were set to 50 and 65, respectively. The total peptide amount normalization method was used to correct for loading bias. S/N (signal-to-noise ratio) was processed through the protein abundance based method to calculate the protein level ratios. The label-free data set was generated as described in the Supporting Information.

Gene Ontology Analysis

First, proteins that were significantly more abundant around either R1 or R2 by comparing receptor-Turbo samples to PM-Turbo-V5 (based on abundance ratios, q < 0.05) were run on the STRING database (String-db.org) as “Multiple proteins” using their UniProt accession numbers, and statistically significant functional enrichments were identified (Figure E,F; Data S2). Next, since these proteomes come from whole cell lysates and are genomic, we ran a ranked functional enrichment analysis (“Proteins with Values/Ranks”, String-db.org) comparing proteomes consisting of all proteins detected and quantitated in at least 2 biological replicates (6164 proteins across all TMT samples, 5779 proteins for all label-free samples) (Figure G,H,K; Data S2). For each comparison, the proteomes were ranked according to their mean raw abundance ratios (e.g., R1^Basal/PM^Basal), and a high (1 percent) FDR stringency was used. , STRING then tests the proteins of each known pathway for any nonrandom skew within the ranked distribution and reports statistically significant pathways based on Kolmogorov–Smirnov and Aggregate Fold-Change testing.

Fly Husbandry

Flies were maintained on standard fly food at 25 °C. UAS-RNAi stocks were obtained from the Vienna Drosophila Stock Center (http://stockcenter.vdrc.at/control/main) and the Bloomington Drosophila Stock Center (https://bdsc.indiana.edu/). We used RNAi lines from three libraries (KK, TRiP, and GD), and all RNAi lines used in this study are listed in Table S1. Each library includes control lines that are in the same genetic background as all the RNAi lines in that library (KK60100 for the KK library, GD60000 for the GD library, and TRIP36304 and TRIP36303 for 2nd chromosome or 3rd chromosome lines from the TRiP library, respectively). For all RNAi experiments, a specific RNAi or their paired control lines were crossed to UAS-Dicer2; R13F02-Gal4. Each experimental day of the screen, progeny from the RNAi line cross are tested alongside the progeny of their corresponding library control cross, and their values within experiment were compared for establishing phenotypes. For behavioral experiments, a mix of males and females was used. For physiological experiments, MB-^Tepac^VV cAMP reporter or UAS-GRAB­(ACh) sensor was used, and only females were imaged due to the fact that their larger size is easier to mount and image.

RNAi Screen Selection

The human Dop1R proximal proteome (Figure B) and ARRB2 (41 proteins) were searched through DIOPT (https://www.flyrnai.org/diopt) for predicted orthologous Drosophila proteins based on sequence. We chose Drosophila orthologs that had a) moderate or high “Rank” on DIOPT, b) measurable expression in any subclass of MBN, and c) “good” quality RNAi (targeting coding or 3′ UTR sequence, all isoforms, and no predicted off-target effects). The RNAi screen consisted of 48 RNAi targeting 48 genes due to more than one Drosophila gene with moderate to high homology for a few human genes.

Aversive Olfactory Paradigm

Flies 1–4 days old were used for standard olfactory aversive conditioning experiments. Aversive olfactory behavioral experiments were conducted under red light and in ∼65–75% humidity at 23–25 °C. Flies were tapped into new food vials and acclimated for 15 min to behavioral testing conditions before training. Groups of ∼70 flies were tapped into tubes where they received 30 s of air, 1 min of CS+ odor paired with electric shock (12 shocks, 1.25 s each, at 5 s intervals, at 90 V), 30 s of air, 1 min of CS– odor with no electric shock, and 30 s of air. After the specified time (3 min or 3 h), flies were loaded into a T-maze and given 1 min to acclimate, then given 2 min to choose between the arms containing CS+ odor and CS– odor. The number of flies in each arm was counted and used to calculate a preference index ((CS−) – (CS+))/((CS−) + (CS+)). Odors used were counterbalanced and included 3-octanol (OCT; 0.05%) and 4-methylcyclohexanol (MCH; 0.07–0.1%) diluted in mineral oil. Odor concentrations were adjusted slightly each day so that flies displayed similar aversion scores to both odors when tested in the T-mazes.

Innate Odor Avoidance Paradigm

Flies 0–3 days old were collected. Fed flies were maintained on normal fly food until testing. Starved flies were moved to vials with only a Kimwipe soaked with 2 mL of deionized water 48–50 h prior to testing. Flies were transferred to a T-maze and given 1 min to acclimate, then given 2 min to choose between the 2 arms, one delivering MCH (0.08%) diluted in mineral oil and the other delivering mineral oil only. A performance index was calculated for each group using the formula: (number of flies in the odor arm – number of flies in the non-odor arm)/total flies in both arms.

Fixed Brain Immunostaining

The brains of 4–6 day-old females were dissected in cold S2 media and immediately placed in 1% paraformaldehyde in S2. Brains were fixed overnight in paraformaldehyde and then washed with PAT3 (0.5% BSA, 1× PBS, 0.5% Triton X-100) for 30 min at a time. Blocking was performed by washing brains with PAT3 + NGS (normal goat serum) for 1 h before primary antibodies. Brains were incubated in the rabbit anti-GFP primary antibody (1:1000) alone (Figure A) or also counterstained with the mouse anti-nc82 antibody (1:50) (Figure A) for 3 h at room temperature and overnight at 4 degrees and secondary antibodies (488 Alexa antirabbit 1:1000 and 633 Alexa anti-mouse 1:1000) for the same duration. Brains were mounted in Vectashield and imaged using a 488 nm laser on a Zeiss Axio Observer Z1 inverted confocal microscope. Images were analyzed by using ImageJ software.

Ex Vivo Functional Imaging

The brains of 4–6-day-old females were mounted flat in a recording chamber, such that the anterior face was facing the microscope objective. Time series images were collected over a period of 8 min at 0.5 Hz. Brains were treated with saline containing 1 μM tetrodotoxin (TTX) 10 min before recording started. Dopamine was applied by switching the solution perfusing the brain from saline to 10 mM dopamine dissolved in saline for 30 s. The levels of cAMP in response to stimulus were reported in terms of the normalized change in inverse FRET ratio (ΔR/R0 = (R – R0)/R0 where R = CFP emission/YFP emission). R0 represents the baseline fluorescence within the ROI. Data were analyzed from stable recordings with no major visible drift using ImageJ software.

Statistical Analysis

For TMT proteomics data, an ANOVA was used to identify proteins that are regulated across comparison groups (R1 or R2 vs PM, Basal vs DA for each receptor, and R1 vs R2 for Basal and DA). The multiple testing correction as per Benjamini–Hochberg (B–H) was applied to identify a “top tier” of significant proteins and limit identification of false positives with a False Discovery Rate (FDR) of 5%. Adjusted p-values, or q-values, of <0.05 were considered statistically significant. For behavioral data, learning and avoidance indices were calculated as described above in “aversive olfactory paradigm” and “innate odor avoidance paradigm” respectively. One-way ANOVA with a Bonferroni post hoc analysis was performed to compare performance indices. For ex vivo cAMP responses to DA, FRET ratios were calculated as described above in “ex vivo functional imaging”. Mann–Whitney U test with Bonferroni correction for multiple comparisons was used. All tests were two-tailed, and confidence levels were set at α = 0.05.

Proximity Labeling of Dop1R1 and Dop1R2 in Cells Using Tail Turbo Fusion

We first chose to use the efficient and live-cell-capable TurboID proximity labeling system by fusing the Turbo ligase to the Drosophila D1-like DA receptors and expressing them in HEK293T cells. This approach takes advantage of the conserved signaling for R1 and R2 between human cells and MBN synapses , to identify candidate memory regulators occupying receptor proximal environments in basal- and ligand-driven signaling and transport. We reasoned that we could identify conserved proteins that we could interrogate in vivo in Drosophila memory circuits.

First, we tested the optimal location for making fusions to R1 and R2 by evaluating the resulting DA-mediated signaling downstream of these receptors using live-cell imaging. We created human cell line expression plasmids (pCAG) containing Dop1R1 (isoform E) and Dop1R2 (isoform A) open reading frames, and we additionally inserted a small chemical tag SNAP (New England Biolabs) with or without a linker (GGS x4) at either the N-terminus or in the cytoplasmic tail of the receptors. To test the resulting impact on receptor function, we utilized live cell optical imaging of cAMP (pGLoSensor, Promega) and intracellular Ca²⁺ (CalfluxVTN) via transient co-transfection of reporter plasmids with R1 (or R2)-SNAP plasmids in HEK293T cells and applied DA (100 μM). For both receptors, N-terminal localization of the SNAP tag, regardless of the presence of a linker, led to a large impairment of DA-mediated cAMP (top) for R1 and R2, and Ca²⁺ (bottom) for R2 (Figure S1A). Notably, tagging the tail with a SNAP tag and a GGS linker had minimal impact on R1 and R2 signaling (cAMP or Ca²⁺), and the tagged receptors were expressed at the plasma membrane (Figure S1B). Therefore, we found that fusions of a tag to the C-terminal tail, including a flexible linker, best maintained the R1/R2 function.

We next tagged the C-terminal (cytoplasmic) tail of R1 or R2 with the intended, but larger, Turbo (+V5 peptide) sequence (Figure A). As a control for ubiquitous Turbo labeling at the plasma membrane and nonspecific Turbo-V5 peptide interactions, we also made a receptor-less Turbo control that included Turbo-V5 targeted to the plasma membrane (PM-Turbo-V5) via Lyn kinase N-terminal peptide. Transient transfection of R1 and R2-Turbo-V5 constructs led to robust expression (via anti-V5 signal) along the plasma membrane, at outstretched protrusions, and internally, in the endomembrane system, compared to empty vector transfection (Figures B and S1C). PM-Turbo-V5 transfection also led to a preference for V5 staining along the plasma membrane and protrusions, similar to R1 and R2-Turbo-V5 expression. Biotin media (500 μM) exposure for 1.5 h led to robust biotinylation of cellular proteins (via Strep-AF488 dye), with overlap particularly at the plasma membrane. Biotin staining internally with no Turbo-V5 overlap is expected due to diffusion after labeling. Furthermore, Turbo-V5 fusion did not disrupt R1-mediated DA→cAMP signaling, but, unlike the smaller SNAP tag, it did reduce DA→R2→Ca²⁺ signaling by ∼50% (Figure C). We noted that comparing V5 staining and biotin-labeled proteomes created by transfection with equal mass of pCAG plasmids containing PM, R1, or R2-Turbo resulted in similar levels of PM and R1-Turbo-V5 expression, but R2-Turbo-V5 levels and resulting Strep labeling were markedly reduced (Figure S1C). It is possible that Turbo fusion to R2 reduced the stability of its mRNA or protein, and this reduced the R2 live-cell signaling we observed in Figure C. To account for this in our proteomics experiment, we adjusted the amount of DNA during transfection for PM, R1, or R2-Turbo-V5 to obtain similar amounts of biotinylated proteome after 40 min of biotin exposure (Figure S1D).

As has been reported, we found that Turbo-mediated biotinylation occurs without biotin supplementation, likely originating from biotin in the standard culture media, but biotin supplementation (500 μM, 1.5 h) robustly increases labeling (Figures B and S1C). Using a shorter time course, we found that biotin addition increased labeling after 10 min and increased further until ∼40 min (Figure S1E,F). Overall, our system allows Turbo labeling of Dop1R environments and can be used to characterize active DA-mediated signaling.

Dop1R Proximal Proteomes Span Receptor Trafficking Compartments

To characterize the basal and ligand-driven protein networks and environments around Dop1R1/2 receptors, we expressed R1-Turbo-V5, R2-TurboV5, or the PM-TurboV5 control, then applied biotin media (500 μM) for 40 min (Figure A). During the 40 min of biotin, we allowed 10 min for biotin to penetrate the receptor environments, a timepoint when Turbo labeling begins (Figure S1E,F), before applying DA for 30 min (“DA”) or not (“Basal”). The cells were lysed, and biotinylated proteomes were bead-purified and digested into peptides. These peptides were processed in one of two methods prior to LC–MS/MS (see Experimental Procedures). For our primary data set, the peptides were labeled with an independent TMT tag per condition (6 tags/6 conditions), and all conditions were then combined for 5 replicates prior to fractionation and LC–MS/MS. As an independent approach, we repeated the experiment with label-free quantitative proteomics that involved no fractionation before LC–MS/MS for all 6 conditions for 2–3 replicates (1 sample was lost). A total of 6164 proteins and 5779 were quantitated across all conditions and replicates in our TMT and label-free data sets, respectively (Figure S2A, Data S1). Despite fundamentally different post-processing of the peptides and fractionation or not, ∼78% of the TMT proteome is detected in the label-free data set. Furthermore, abundance ratios (abs­(log2 ratio) > 0.5) between conditions (e.g., R1 Basal/PM Basal) for TMT proteins were positively and linearly correlated with their abundance ratio in the label-free data (Figure S2C), indicating remarkable reproducibility of data sets. However, given the increased quantitative power of TMT proteomics and its larger sample size, going forward, we use the label-free data as support for findings from the more statistically robust TMT data set.

Using standard multi-test correction q-value analysis on our TMT data set, we found 35 human proteins were significantly more abundant in samples (basal and DA) from R1-Turbo (29 proteins) or R2-Turbo (25 proteins) compared to the receptor-less plasma membrane bound PM-Turbo control (q < 0.05, Figure B), with 10, or 6 proteins, uniquely significant for R1 and R2, respectively (Figure B, Figure S2D, Data S1). 11 human proteins had R1/R2 abundance ratios significantly different, all of which are biased to be labeled by R1-Turbo (Figure C). Many of the 11 R1-specific proteins are also significantly more abundant in R1-expressing samples compared to PM control (Figure B). Importantly, the Drosophila proteins Dop1R1 and Dop1R2 themselves were robustly present only under conditions where the respective receptor was expressed and at similar levels between basal and DA conditions (Figure S2E), indicating Dop1R-Turbo self-labeling, validating our Turbo methodology. Remarkably, most of the TMT significant proximal proteins also have similar abundance ratios between TMT and label-free data sets, as seen on volcano plots (Figure S2F,G). Gene ontology (GO) analysis (String.db) on this significant Dop1R proximal network shows significant enrichment for proteins found in cellular locations expected along the transport/lifespan of these GPCRs, including the ER, Golgi, endosome, and PM (Figure D,E, Data S2). This further validates that we indeed captured Dop1R proxiomes and potential interacting proteins.

R1-Turbo and R2-Turbo Are Localized to Distinct Subcellular Environments

R1 and R2 drive distinct cellular signaling in human cells and synapses, , indicating that the receptors are located in differential signaling environments and subcellular locations. We first compared R1 and R2-Turbo proteomes using GO analysis with the small 11 protein network that is significantly (q < 0.05) more abundant around R1 than R2. Given the small network, we did not find significant enrichments in traditional GO categories, but we did find two significant enrichments for UniProt Keywords “Cytoskeleton” and “Endocytosis”, suggesting that R1 is proximal to pathways involving these keywords (Figure F). To fully utilize our two data sets and gain more insights into differences between R1 and R2 functions, we performed a ranked functional enrichment analysis (String-db.org) using all quantified proteins (no cutoffs) ranked by raw abundance ratios (6164 proteins for TMT, 5779 for label-free). This analysis utilizes our large whole cell-derived data set to find statistically significant pathways by testing the proteins of each known pathway for any nonrandom skew within the ranked distribution. We used our data sets ranked by R1/R2 ratios (Figure G,H, Data S2) to find the top significant enrichments. We found many R1-biased pathways that included actin regulation, clathrin, synaptic vesicle endocytosis, or were located at the early endosome membrane for both data sets (Figure G, Data S2). This corroborates our 11-protein q-value analysis on the cytoskeleton and endocytosis pathways above (Figure F). Many proteins in these pathways have positive R1/R2 abundance ratios (Top R1/R2 basal ratio proteins, Figure I) and large average normalized abundance around R1-Turbo compared to R2-Turbo and PM-Turbo (Figure S2H). This pattern is remarkably conserved across TMT and label-free data. In particular, Clathrin-based endocytosis-regulating proteins FNBP1 and SYNRG are significantly more abundant in R1 than R2 samples (Figure C) as well as EPS15 and EPS15L1 (not significant) have large R1/R2 ratios in both data sets (Figures I and S2H). Furthermore, WASHC2A, a WASH complex protein involved in endosomal sorting and recycling back to the plasma membrane, was significant in both R1 vs PM and R1 vs R2 q-value analyses (Figure B,C) and is significantly more abundant in R1 samples than R2 (Figures C,I and S2H). While WASHC2C, a second WASH complex protein, did not reach q-value significance, it was clearly enriched in R1 vs R2 (Figures I and S2H). This suggests R1 might be sorted and recycled in the endosomal system differently than R2. Overall, R1 appears to be regulated at the membrane via canonical clathrin-mediated synaptic vesicle endocytosis, and R1’s proximity to cytoskeleton regulatory proteins, at least in part, reflects the importance of the cytoskeleton in vesicle endocytosis and sorting in the endosome.

Despite R2-Turbo localization at the plasma membrane (Figures B and S1B,C), proteins related to synaptic vesicle and clathrin-mediated endocytosis were labeled far less by R2-Turbo (Figures I and S2H), suggesting R2 might be internalized differently from R1. Interestingly, the top significant enrichments with an R2 bias included integrins, cell adhesion, and postsynaptic density environments for both data sets (Figures H and S2I, Data S2). We noted that the top 2 postsynaptic density proteins with R2 bias were Ephrin type-B receptor 2 (Ephb2) and neuroplastin (Nptn) (Figure S2I). Both Ephrin signaling and Nptn mediate cell–cell signaling that plays a role in synaptic plasticity underlying memory in mammals and Drosophila. − We also observed bias toward labeling nuclear chromosome-associated proteins, like histone H2AX, with R2-Turbo compared to R1-Turbo (Figures H and S2I); however, this was not reproduced in the label-free data set. Whether this indicates R2’s localization near the nucleus or nuclear proteins’ localization near the plasma membrane remains unclear. Overall, these results indicate R1-Turbo and R2-Turbo fusion proteins are localized in separate subcellular compartments near or at the plasma membrane, which could support their distinct cellular functions in synapses.

DA Differentially Shifts R1 and R2 Signaling Environments

Ligand binding to a GPCR usually drives its subsequent internalization (or endocytosis) away from the plasma membrane, and this desensitizes the cell’s response to subsequent ligands, , and this has been seen to occur in minutes, and proximity labeling has been used to observe this movement for other receptors. Desensitization involves phosphorylation by G-protein receptor kinases (GRKs), binding to β-Arrestin2, internalization, and movement of the receptor to the early endosome. , Thus, we aimed to use our Dop1R-Turbo system to capture the movement of these receptors from “Basal” environments to DA-bound environments (Figure J). Comparing ligand (DA)-driven and basal proteomes within the TMT data set revealed no protein reached significance with our strict q < 0.05 cutoff for any Turbo condition. We ran a ranked functional enrichment analysis on all quantitated proteins ranked by raw abundance ratio (DA/basal). Proteins with high DA/basal ratios should be enriched for new signaling environments driven by ligand activation (DA bias), and those with low DA/basal ratios would be enriched for old signaling environments left behind after ligand activation (Basal bias). We found that many GO terms were significantly enriched with either Basal or DA bias and were both common and unique for both receptors for the TMT data set (Figure K; Data S2).

Notably, we found that for R1-Turbo, the top significantly enriched pathways with a Basal bias included terms involving anchored components of the membrane and GPI-anchoring (Figure K, left). Many proteins in the TMT data set associated with these pathways had reduced abundance in the presence of DA for R1-Turbo and, to a lesser extent, the label-free data set (Figure S2J). In particular, SMPDL3B, PRNP, CD44, and CD55, proteins targeted to lipid rafts via GPI-anchor, were markedly reduced in R1 samples when DA is present (Figure S2J), suggesting R1 resided in lipid raft microdomains in the plasma membrane before being internalized after activation.

Interestingly, we saw a Basal bias for R1-Turbo samples for proteins belonging to inhibitory G alpha pathways, including the terms Gaz and Gai signaling events (Figure K, left). This Basal bias for R1-Turbo is mostly attributable to a variety of GPCR signaling proteins, including the adenylyl cyclase ADCY5, the G protein gamma subunit GNG12, the inhibitory G alpha GNAI2, and others (Figure S2J, “Ga­(i/z) signaling”). We also looked at all measured G alpha subunits in our data sets, and, other than GNAI2, we did not see robust changes in abundance between Basal and DA conditions for either receptor or between receptors (Figure S2J, “G alpha measured”).

Regarding R2 and using the TMT data set, we found the top significantly enriched pathways with a Basal bias included similar pathways as those with R2 bias (Figure H), specifically ephrin signaling, postsynaptic density, and cell adhesion pathways (Figure K, left, purple bars). Therefore, unlike R1-Turbo, DA drives R2-Turbo away from cell–cell adhesion-like signaling environments at the membrane.

On the other hand, the top significantly enriched terms for R1 with a DA bias in the TMT data set included a STRING cluster related to GPCR desensitization via ARRB2 and GRKs (Figure K, right) that includes ARRB2 (Beta-arrestin 2), GRK2, and GRK3, all of which have large positive DA/Basal ratios (Figure L, top) and increased abundance in R1-Turbo samples exposed to DA (Figure L, bottom), and this is conserved across data sets. Interestingly, ARRB2 was robustly enriched (∼2-fold) when either R1-Turbo or R2-Turbo receptors and DA were simultaneously present but low in all other conditions for the TMT data set (Figure M, top-left). In the label-free data set, ARRB2 was significantly increased for both receptors in the presence of DA compared to PM-Turbo + DA (Figure M, bottom-left). Furthermore, we found that in the label-free data set, GRK3 abundance was significantly lower in basal condition for R1-Turbo samples compared to both PM and R2-Turbo samples, and DA drove a trending, but not significant (#, p = 0.059), increase in normalized abundance for GRK3 only in R1-Turbo samples (Figure M, bottom-right). Given ARRB2 and GRK3’s direct role in ligand-driven endocytosis of GPCRs, these results argue we indeed captured DA-driven receptor interactions. Furthermore, while ARRB2 was recruited by both R1-Turbo and R2-Turbo, they might have differential engagement with GRKs when activated by DA.

Additionally, the pathways of clathrin-coated pits, presynaptic endocytosis, and early endosome membrane were significantly enriched when DA is present for R1-Turbo (but not R2-Turbo) (Figure K, right), with many proteins having high DA/basal ratios in R1-Turbo samples (Figure N), and thus increasing their abundance in the presence of DA in R1-Turbo samples (Figure S2K). Many of these proteins had similar patterns in the label-free data set for early endosome membrane, but not for clathrin endocytosis. Altogether, these data indicate that Turbo labeling captured DA-mediated shifts in R1’s location away from the plasma membrane and toward clathrin-mediated internalization pathways and to the early endosome.

Looking at the DA bias regarding R2, we found, similar to R1, a significant enrichment for the early endosome membrane with a DA bias. Therefore, this result, combined with our findings that DA drives R2’s proximity/interaction with ARRB2 (Figure K), indicates that DA drives R2-Turbo internalization away from the membrane to the early endosome, but possibly not through clathrin-mediated endocytosis like R1.

Finally, we found that in the TMT data set, both R1 and R2 show significant enrichment for “structural constituents of chromatin” with a DA bias, and these consist mostly of histones (Figures K and S2L), like “nuclear chromosome” proteins with R2 bias in Figure S2I. This was again unexpected and could indicate that histones are moonlighting outside of the nucleus near the plasma membrane, or ligand activation of these dopamine receptors drives localization inside the nucleus. , However, we did not replicate this DA-mediated increase in abundance of these chromatin proteins in the label-free data set. Nevertheless, our results indicate that our proximity labeling approach was successful in capturing ligand-driven shifts in receptor location and signaling, suggesting fundamental differences between these receptors in their plasma membrane localization (GPI anchor vs cell adhesion) in basal conditions, DA-driven engagement with GRKs, and endocytosis pathway (clathrin or not) when activated by DA.

Trafficking Proteins Identified to Play a Role in Two Dopamine-Mediated Behaviors

Given that R1 was proximal to cell components and pathways related to vesicle and GPCR trafficking for both data sets (Figures I and S2, Data S2), we sought to identify R1 trafficking proteins that regulate DA-mediated learning in Drosophila. First, we targeted the Drosophila orthologs of the human Dop1R proximal network (39 proteins, q < 0.05, Figure B) and ARRB2 in the MBN using the R13F02-Gal4 line commonly used in the field for MBN preferential expression (Figure A). ,,,,, We crossed R13F02-Gal4 (with UAS-Dicer2) to drive RNAi expression and cause gene-specific RNA interference in the MBN (Figure B), and we tested R1-dependent aversive odor learning (Figure C). During training, animals were first exposed to one of two odors (3-Octanol, OCT or Methylcyclohexanol, MCH) simultaneously with 6 electric shocks (moderate training), followed by exposure to the other odor. To test learning, we tested the memory immediately after training. Disruption of many orthologues of Dop1R proximal proteins tended to enhance or disrupt immediate memory (Figure D), as their mean ± SE did not overlap with the mean ± SE of the same-day control (#). Among these, we noted disruption of two proteins related to ER→Golgi transport, Sec24AB and CG13887, which disrupted and enhanced memory, respectively. Furthermore, disruption of krz, an ortholog of human ARRB2 and likely a regulator of endocytosis, tended to reduce learning. Similar to our proteomics results for ARRB2 (Figure K), human orthologs for Sec24AB (human Sec24B) and CG13887 (human BCAP31) showed significant enrichment in R1 and R2-Turbo-V5 samples regardless of DA application in the TMT data set (and also for BCAP31 in the label-free data set as well) (Figure S3A).

To validate these potential R1 trafficking proteins’ role in learning, we targeted Sec24AB, CG13887, and Krz in the MBN using the screen RNAi line and an additional RNAi line targeting an independent portion of each gene. Disruption of Sec24AB and Krz in the MBN significantly decreased immediate memory with both RNAi lines compared to two genetic controls (1 lacking Gal4, 1 lacking RNAi) (Figure E,F). Remarkably, disruption of CG13887 in the MBN significantly enhanced immediate memory for both RNAi lines compared to controls (Figure G,G′). Finally, disruption of these genes in the MBN did not impair odor or shock avoidance behavior, with the exception that odor avoidance to only one odor (MCH) was slightly but significantly reduced for Krz (Figure S3B,C).

To evaluate how these genes regulate the acquisition of associative memory across training intensity, we varied the number of electric shock punishments coinciding with the odor. We found that disruption of Sec24AB in the MBN did not disrupt immediate memory at the lowest training intensity (1 electric shock) but did significantly decrease immediate memory with 2 electric shocks and above (Figure S3D, top), suggesting Sec24AB is required for strengthening the association with additional punishments but not required for the initial association. Disruption of Krz in the MBN significantly decreased only the highest intensity training (Figure S3D, middle), indicating that it is required for maximal memory strength. Finally, disruption of CG13887 enhanced immediate memory at 6 and 12 electric shock associations, suggesting CG13887 normally limits the maximal strength of the association and is a memory suppressor gene.

Krz is one of two nonvisual arrestins in Drosophila, alongside CG32683. We targeted CG32683 in the MBN and did not see a significant decrease in learning (Figure S3E). This suggests that Krz might be the primary nonvisual arrestin supporting learning. Sec24AB, and its paralog Sec24CD, in flies are critical components of the COPII pathway and predicted to be two adaptor proteins that recognize distinct GPCR cargo for ER export, alongside cargo-invariant COPII components like Sec16 and Sar1. Interestingly, for almost all COPII components (Sec16, Sar1, and Sec24CD), disruption with two independent RNAi lines led to lethality of progeny, making memory analysis impossible. Only one Sec16 RNAi line survived, but this line showed no impact on learning (Figure S3F). This suggests that the COPII pathway (and transport of GPCRs generally) is required in MBN (or elsewhere within R13F02-Gal4 expression) for survival for unclear reasons. Sec24AB is not required in the MBN for survival and perhaps has a specific role in transporting learning-related GPCRs like R1.

If these proteins traffic R1 to regulate learning, then we anticipate that the disruption of these trafficking proteins would alter other R1-mediated behaviors. Recently, it was shown that hungry flies have increased attraction to novel odors, and this requires R1 in MBN synapses. R1 signals via distinct adenylyl cyclases, resulting in distinct types of synaptic plasticity compared to aversive learning. In this assay, adult animals are either given water only (starved) or food (fed) for 50 h and then tested for their preference between a novel odor (MCH) and no odor (Figure H). As shown previously, normally fed flies (with or without R1 disruption) avoid novel odors at the concentrations used in this assay (Figure I). This avoidance causes flies to have large negative odor-attraction values. However, when control flies are starved, they become significantly less averse (more attracted) to the novel odor MCH, but if R1 is disrupted in the MBN, this significant shift in odor attraction does not occur, as reported previously. Remarkably, and similar to R1, disruption of Krz and CG13887 in the MBN prevented the starvation-induced shift in odor attraction but not Sec24AB (Figure J). Altogether, we have identified three trafficking genes orthologous to members of our Dop1R1 proximal proteome that regulate two distinct R1-mediated pathways and behaviors (Krz and CG13887) or just learning (Sec24AB).

Trafficking Proteins Sec24AB, CG13887, and Krz Regulate DA Signaling in MBN

To test whether these trafficking proteins that modulate R1-based behaviors alter the overall levels of R1 in the MBN, we imaged and immunostained brains expressing Dop1R1 endogenously tagged with mVenus and coexpressing RNAi for Sec24AB, Krz, CG13887, or no RNAi control in the MBNs and quantified the levels of R1. We found that the R1 abundance in the mushroom body lobes, as well as the fan-shaped body and central brain overall, was not affected by the MBN expression of RNAi for Sec24AB, Krz, or CG13887 (Figure A,B). These data indicate that the general production and transport of R1 out to the synaptic neuropil are not affected by these trafficking proteins. However, short-range trafficking of R1 to and from the plasma membrane at the synapse would not be measurable on this gross anatomical level.

To more directly test the role of these trafficking proteins on DA→R1-mediated synaptic signaling in the MBN, we conducted ex vivo imaging of downstream cAMP response to DA perfusion across the brain using an MBN-driven cAMP FRET sensor (MB-Epac). We reasoned that if the disruption of these trafficking proteins altered the presence of R1 at the membrane, we should see altered DA-mediated cAMP in the MBN neuropil (Figure C). We measured cAMP responses (change in FRET ratio YFP/CFP compared to baseline) to a 30 s perfusion of 10 mM DA at the γ2 compartment of the MBN, a synaptic neuropil critical in the acquisition and expression of short-term aversive memory ,− and starved odor avoidance behavior (Figure D). As expected, direct disruption of R1 signaling through treatment with the D1-like receptor inhibitor SCH23390, or R1 RNAi expression in the MBN, significantly impaired DA-mediated cAMP responses (Figure E,F) in terms of total (Figure E′,F′) and peak responses (Figure E″,F″’). These experiments confirm that cAMP response to DA is highly dependent on R1.

Finally, we disrupted Sec24AB, Krz, and CG13887 trafficking proteins in the MBN and found that Sec24AB RNAi expression caused a trending, but not significant, disruption in overall and peak cAMP responses to DA (Figure G–G″), whereas Krz RNAi expression significantly reduced total cAMP responses to DA, but not the peak (Figure H–H″). On the other hand, CG13887 RNAi expression in the MBN significantly enhanced peak cAMP responses to DA, but not the total (Figure I–I″). These physiological data suggest that Sec24AB and Krz are required for high levels of cAMP in response to DA, whereas CG13887 normally suppresses cAMP response to DA, thus mirroring their respective effects on learning (Figure E–G).