Abstract

   Transmembrane proteins play vital roles in mediating synaptic
   transmission, plasticity, and homeostasis in the brain. However, these
   proteins, especially the G protein–coupled receptors (GPCRs), are
   underrepresented in most large-scale proteomic surveys. Here, we
   present a new proteomic approach aided by deep learning models for
   comprehensive profiling of transmembrane protein families in multiple
   mouse brain regions. Our multiregional proteome profiling highlights
   the considerable discrepancy between messenger RNA and protein
   distribution, especially for region-enriched GPCRs, and predicts an
   endogenous GPCR interaction network in the brain. Furthermore, our new
   approach reveals the transmembrane proteome remodeling landscape in the
   brain of a mouse depression model, which led to the identification of
   two previously unknown GPCR regulators of depressive-like behaviors.
   Our study provides an enabling technology and rich data resource to
   expand the understanding of transmembrane proteome organization and
   dynamics in the brain and accelerate the discovery of potential
   therapeutic targets for depression treatment.

INTRODUCTION

   As the most complex organ of the mammalian body, the brain has been
   intensively characterized at the molecular level in a system-wide
   fashion using a variety of transcriptomic or imaging approaches. The
   Allen Brain Atlas ([63]https://portal.brain-map.org/) hosts a plethora
   of in situ hybridization (ISH) and microarray-based databases to
   describe the regional or cellular gene expression profiles of adult and
   developing mammalian brains ([64]1–[65]4). The Human Protein Atlas
   program ([66]www.proteinatlas.org), together with a recent brain atlas
   project, has concertedly established high-quality transcriptome and
   protein imaging resources to map the spatial expression of transcripts
   and proteins across multiple mammalian brain regions ([67]5, [68]6).
   Moreover, given that all the functions of the brain are ultimately
   mediated by proteins and that poor correlation between mRNA and protein
   abundances has been observed in various cell types and tissues
   ([69]7–[70]9), large-scale mass spectrometry (MS)–based proteomic
   surveys have been launched to map protein expression patterns in an
   unbiased manner across multiple regions of mouse and human brains
   ([71]10, [72]11). However, despite our increasing ability to
   interrogate the molecular organization of the brain, in-depth and
   quantitative profiling of transmembrane protein expression is a notable
   exception.

   G protein–coupled receptors (GPCRs), ion channels, and transporters
   constitute three prominent cell-surface transmembrane protein families
   that play essential roles in mediating neuronal signal processing and
   plasticity in the brain ([73]12–[74]14). A number of their family
   members, especially GPCRs, represent the most successful targets of
   molecular therapeutics for central nervous system (CNS) disorders
   ([75]12, [76]15). However, these transmembrane proteins are especially
   challenging to measure using conventional proteomics techniques owing
   to their strong hydrophobicity, relatively low abundance, and fast
   turnover ([77]8). Notably, GPCRs are notoriously underrepresented in
   current MS-based proteomic surveys. For example, a proteome atlas of 29
   healthy human tissues profiled 103 GPCRs in total, 64 of which were
   from brain tissue of the 831 GPCRs encoded in the human genome ([78]8).
   A similarly low coverage of GPCR identification (ID) was reported in
   another global proteomic analysis of human cells (56 GPCRs among 14,237
   identified proteins) ([79]16).

   Here, we present a new proteomic approach for deep and accurate
   profiling of low-abundance transmembrane protein families in the
   region-resolved mouse brain. Our approach integrates three innovations
   compared to conventional proteomic workflows. First, we performed cell
   membrane fractionation to reduce the abundant cytosolic proteins and
   enrich transmembrane proteins. Second, we carried out single-shot
   data-independent acquisition (DIA) MS analysis rather than conventional
   data-dependent acquisition (DDA) analysis as DIA MS is an emerging
   technology with superior accuracy and reproducibility in proteomic
   quantification ([80]17, [81]18). Last, we created a GPCR
   family–targeted hybrid library using deep learning tools for DIA MS
   data mining so as to achieve an unprecedented depth of transmembrane
   protein profiling. Using this approach, we were able to identify and
   quantify 143 GPCRs, 170 ion channels, and 176 transporter proteins
   across 10 mouse brain regions.

   By matching our multiregional proteomics profiling data with the
   genome-wide transcriptomics and ISH data, we identified region-enriched
   GPCRs and other transmembrane proteins with considerably discordant
   mRNA and protein distribution over multiple brain regions. Through
   protein coexpression analysis, we predicted an endogenous GPCR
   interaction network in the mouse brain and validated the colocalization
   of a GPCR protein and its unknown interacting partner in neuronal cell
   culture and brain tissue. Furthermore, we used this new workflow to
   reveal the landscape of transmembrane proteome remodeling in 11 mouse
   brain regions of a chronic stress–induced depression model, which led
   to the rapid discovery of two novel GPCR regulators of depressive-like
   behaviors.

RESULTS

Profiling the transmembrane proteome in multiple mouse brain regions by
single-shot DIA MS analysis

   To profile transmembrane protein expression in the region-resolved
   mouse brain, we collected 10 anatomically dissected adult mouse brain
   regions. Nine and eight regions overlap with those documented in the
   Allen Mouse Brain Atlas and the recently published brain transcriptome
   atlas ([82]6), respectively ([83]Fig. 1A). To increase the proteome
   coverage of transmembrane proteins, we isolated cell membrane
   fractions, performed membrane protein extraction and digestion under
   optimal conditions, and analyzed protein digests from individual brain
   regions with single-shot DIA MS ([84]Fig. 1B and fig. S1A). For each
   brain region, we also fractionated the pooled replicates and performed
   DDA MS analysis of prefractionated samples to build a project-specific
   spectral library for DIA MS data mining ([85]Fig. 1B, left).

Fig. 1. Transmembrane proteome profiling of the region-resolved mouse brain
with DIA MS analysis.

   Fig. 1
   [86]Open in a new tab

   (A) A summary of the brain regions of the resting-state mice examined
   in this study (left) and the overlapping regions analyzed using ISH by
   Allen Brain Atlas or RNA-seq by Mulder and colleagues ([87]6) (right).
   (B) Overall workflow of mouse brain DIA MS analysis and data mining.
   For each brain region, membrane proteins [represented by three
   transmembrane protein families this study focuses on] were isolated,
   extracted, and digested before MS analysis (top). A project-specific
   DDA library was built from DDA MS analysis of prefractionated
   multiregional brain tissues (left route). Meanwhile, a GPCR
   family–targeted hybrid library was built by merging an initial DIA
   library derived from the DIA MS data and a GPCR virtual library
   predicted from 524 mouse genome–encoded GPCR sequences using deep
   learning models (right route). The DIA MS data were searched with the
   two libraries to yield transmembrane protein ID and quantification
   results.

   This DDA experiment–derived spectral library (DDA library for short)
   comprised a total of 134,560 peptide precursors mapped to 7995 protein
   groups, of which 2510 were transmembrane proteins (with at least one
   transmembrane domain). Mouse genome encodes 524 GPCRs, 316 ion
   channels, and 296 transporters (protein lists in table S1). Among them,
   135 GPCRs, 207 ion channels, and 207 transporters were present in the
   DDA library ([88]Fig. 2A and fig. S1B). Although the proteome coverages
   of three transmembrane protein families were lower than those detected
   at the transcript level by ISH or RNA sequencing (RNA-seq), our results
   significantly outnumber the most comprehensive mouse brain proteomic
   survey reported to date (fig. S1B) ([89]10).

Fig. 2. Deep proteome coverage and reproducible quantification of GPCR, ion
channel, and transporter family members in the mouse brain.

   Fig. 2
   [90]Open in a new tab

   (A) Number of protein IDs in the DDA library and the GPCR hybrid
   library. The full complement of 524 genome-encoded GPCRs are included
   in the latter. (B) Number of GPCR IDs in each brain region yielded with
   the DDA library (purple) or the GPCR hybrid library after data
   filtering (light purple, shared IDs between the two libraries; orange,
   unique IDs only yielded with the hybrid library). OLF, olfactory bulb;
   CBC, cerebral cortex; CB, cerebellum; HIP, hippocampus; MB, midbrain;
   SC, spinal cord; STR, striatum; TH, thalamus; PO, pons; MY, medulla.
   (C) Comparison of GPCR IDs from 10 brain regions yielded with the two
   libraries. (D) Number of protein IDs for three families in each region
   and in total. GPCR IDs are concatenated from two libraries, and ion
   channel and transporter IDs were detected with the DDA library. (E)
   Subcellular localization of all protein IDs according to gene ontology
   cellular component classification. ER, endoplasmic reticulum; Mt,
   mitochondria. (F) Many enriched biological processes (P < 10^−5) in all
   protein IDs are related to neuronal cell activity or brain functions.
   (G) Spearman correlation of protein quantification between replicates
   of each region.

Deepening the GPCR subproteome coverage with a targeted hybrid library
strategy

   Given that GPCRs are particularly challenging to map with conventional
   proteomics techniques ([91]8), we developed a targeted hybrid library
   strategy to deepen the coverage of a selected transmembrane protein
   family (see Materials and Methods for details). Briefly, we created a
   GPCR family–targeted hybrid library (GPCR hybrid library for short)
   using deep learning models ([92]Fig. 1B, right) ([93]19, [94]20), which
   contains the full complement of 524 GPCRs encoded in the mouse genome
   ([95]Fig. 2A). To control the false discovery rate (FDR) when using a
   targeted hybrid library, we implemented an additional data filtering
   criteria (Cscore, >1.0) to restrict the subgroup FDR of GPCR peptide ID
   as assessed using a decoy library approach (fig. S2, A to C). This
   strategy was applied to processing our 10–brain region DIA MS data.

   An average of 108 GPCR proteins were identified per region using the
   GPCR hybrid library after data filtering, whereas only an average of 65
   GPCR proteins were identified with the DDA library ([96]Fig. 2B).
   Moreover, we observed similarly high quantification reproducibility for
   all proteins identified with the two libraries (fig. S2D). By combining
   GPCR IDs from two libraries, we substantially increased the GPCR
   subproteome coverage in the mouse brain through single-shot DIA MS
   analysis. Of the 143 concatenated GPCR IDs, 56 were exclusively
   detected using the targeted hybrid library strategy ([97]Fig. 2C).
   Although we previously demonstrated this strategy in processing data
   from a few brain regions as a proof of a concept ([98]21), our current
   study proved that this hybrid spectral library can deepen the GPCR
   subproteome coverage to surpass a large-scale project-specific DDA
   library.

   In addition to GPCRs, our transmembrane proteome profiling identified
   and quantified 170 ion channels and 176 transporters across 10 brain
   regions ([99]Fig. 2D). Of all protein groups profiled in our study,
   58.5% are located in plasma membrane, intracellular membrane, or
   synapse ([100]Fig. 2E). Most of them are enriched in biological
   processes closely related to neuronal cell activity or brain functions,
   such as neuron projection development, neurotransmitter transport,
   sensory perception of pain, and fear response that are known to be
   mediated by the three transmembrane protein families ([101]Fig. 2F). In
   regard to the quantification performance of our new workflow, all
   proteins profiled between independent replicates of each brain region
   showed strong correlation and low quantification deviation [median
   coefficient of variation (CV), 8.3%], which indicates superior
   quantification consistency by our DIA MS analysis over the previous
   mouse brain proteomic analysis (median CV, 28.1%) ([102]Fig. 2G and
   fig. S2E) ([103]10).

Comparison of the transcriptome and proteome profiles for transmembrane
proteins

   Our transmembrane proteome profiling enables a global view of
   transmembrane protein expression across different mouse brain regions
   that can be compared with region-resolved gene expression at the
   transcriptome level (table S2). The principal components analysis (PCA)
   and the hierarchical clustering tree based on our quantification of all
   transmembrane proteins revealed four clusters of brain regions:
   pons/medulla, spinal cord/midbrain, cerebral cortex/striatum/thalamus,
   and olfactory bulb/hippocampus (HIP)/cerebellum ([104]Fig. 3A). In
   contrast, the transcriptome profiles of transmembrane protein–coding
   genes showed a different pattern of regional connectivity, with
   cerebral cortex/HIP/amygdala being clustered more tightly and
   cerebellum standing out as an outlier ([105]Fig. 3B).

Fig. 3. Multiregional expression profiles of transmembrane protein–coding
genes at the transcriptome and proteome levels.

   Fig. 3
   [106]Open in a new tab

   (A and B) PCA (top) and HCA (bottom) showed different patterns of
   regional expression of transmembrane protein–coding genes based on
   proteomics data (this study) (A) versus RNA-seq data ([107]6) (B). (C)
   Spearman correlation of expression profiles across at least five brain
   regions between the proteomic and RNA-seq measurement for all
   transmembrane (TM) proteins and three transmembrane protein families.
   Median correlation coefficients are shown above the plots. (D) Examples
   of discordant expression profiles measured by proteomics, qPCR, and
   ISH. (E) Validation of regionally elevated expression of transmembrane
   proteins with inconsistent mRNA profiles by immunoblotting. (F) ID of
   regionally elevated GPCRs, ion channels, and transporters in the three
   categories (region-enriched, group-enriched, and enhanced). Total
   numbers of elevated proteins in each region are provided to the right.
   (G) Spearman correlation of expression profiles across at least five
   brain regions indicated the lowest protein-to-mRNA correlation for
   region/group-enriched GPCRs. Median correlation coefficients are shown
   above the plots. (H and I) Relative expression profiles of
   region/group-enriched GPCRs with negative mRNA-to-protein correlation
   (H) and region/group-enriched orphan receptors (I). The correlation
   coefficient of proteomic versus RNA-seq or proteomic versus ISH
   measurement is annotated for each GPCR. No correlation available for
   three receptors of which the expression profiles were overlapped
   between less than five regions.

   An overall modest correlation between mRNA and protein abundances was
   observed for 1738 transmembrane protein–coding genes (median Spearman
   correlation coefficient, 0.39) shared between the RNA-seq and
   proteomics data (table S3). Among them, 392 genes (22%) showed negative
   correlation, indicating substantial difference in their mRNA and
   protein expression patterns ([108]Fig. 3C). We also analyzed 1733
   transmembrane protein–coding genes shared between the ISH and
   proteomics data and found an even weaker correlation (median
   correlation coefficient, 0.22; 32% genes showing negative correlation)
   (fig. S3A and table S3). Moreover, GPCR and transporter family members
   showed lower mRNA-to-protein correlation than ion channels ([109]Fig.
   3C and fig. S3A).

   To validate the distinct interregional protein expression profiles
   revealed by our proteomics analysis, we selected one ion channel [Gria2
   (glutamate receptor 2)] and two transporters [Slc6a3 (sodium-dependent
   dopamine transporter) and Slc5a7] with inconsistent RNA expression
   profiles to be examined by immunoblotting and immunostaining. Slc6a3
   protein was predominantly expressed in the striatum, yet its mRNA was
   mainly detected in the midbrain and pons by quantitative polymerase
   chain reaction (qPCR) and ISH ([110]Fig. 3D). Immunoblotting showed
   almost exclusive presence of Slc6a3 protein in the striatum ([111]Fig.
   3E). Immunostaining of the brain slice confirmed this result and
   further revealed the distribution of Slc6a3 in the axons of projection
   neurons (fig. S3, C to E). For Gria2 and Slc5a7, their protein
   expression patterns were validated to be appreciably different from
   their mRNA expression ([112]Fig. 3, D and E). Therefore, our DIA
   MS-based proteome profiling provides a high-throughput and accurate
   measure of transmembrane protein distribution, which, in many cases, is
   largely discordant with the transcript distribution.

Brain region–enriched GPCRs and other transmembrane proteins

   Using the region-averaged proteomic quantification data (table S4), we
   identified regionally elevated GPCR, ion channel, and transporter
   family members that were classified into three categories:
   region-enriched proteins (twofold higher abundance than all other brain
   regions), group-enriched proteins (two to four brain regions with
   twofold higher abundance than all other regions), and enhanced proteins
   (twofold higher abundance than the median of all other brain regions)
   ([113]Fig. 3F and table S5). The GPCR family contains the largest
   number of regionally elevated proteins (three categories together)
   ([114]Fig. 3F), in accordance with the largest variation of
   interregional protein abundances observed for GPCRs (fig. S3F). The
   striatum has the most region-enriched GPCRs, whereas most
   group-enriched GPCRs are shared among the olfactory bulb, midbrain, and
   HIP (table S5). Notably, the interregional mRNA-to-protein expression
   correlation was much weaker for region/group-enriched GPCRs than for
   region/group-enriched ion channels or transporters ([115]Fig. 3G and
   fig. S3B).

   Our study suggests that this subset of GPCRs enriched in certain brain
   regions experience unusually pronounced regulation of protein
   synthesis, degradation, or transport. This is exemplified by 11
   region/group-enriched GPCRs with negative mRNA-to-protein correlation,
   including members from the chemokine receptor, adrenergic receptor, and
   neuropeptide Y receptor families ([116]Fig. 3H). We also analyzed 15
   region/group-enriched orphan GPCRs for which native ligands are unknown
   and physiological functions are largely unexplored ([117]22, [118]23).
   The multiregional protein distribution for eight enriched orphan GPCRs
   were negatively correlated with their mRNA distribution ([119]Fig. 3I).
   For instance, Gpr161 gene transcription mainly occurred in the HIP and
   olfactory bulb, but most of its protein product was likely to be
   transported to the cerebral cortex in addition to the olfactory bulb.
   The extensive efferent projections from the HIP and olfactory bulb to
   the prefrontal, cingulate, retrosplenial, and the olfactory cortex
   ([120]24–[121]26) might explain the enriched expression of Gpr161
   protein in the cerebral cortex. For Gpr101, Gpr34, and Gpr62, their
   transcripts were distributed evenly across multiple regions, yet the
   protein product was highly enriched in one specific region. Together,
   the protein distribution patterns of these brain region–enriched GPCRs
   would shed new light on their posttranscriptional regulation and
   uncharacterized functions in the brain.

GPCR interaction prediction based on multiregional protein coexpression
analysis

   Protein coexpression or coregulation analysis based on the quantitative
   proteome profiling data can be exploited to infer the composition of
   protein complexes and their interaction networks ([122]27, [123]28).
   Thus, we reasoned that it may be possible to find unknown
   protein-protein interactions (PPIs) from our multiregional proteomics
   resource. As a proof of concept, for each measured GPCR in our dataset,
   we extracted their potential interacting partners with correlated
   expression profiles over at least five brain regions. Using a commonly
   applied cutoff for positive correlation [Pearson correlation
   coefficient (PCC), >0.7] ([124]27, [125]28), we initially identified
   16,074 potential PPIs for 124 GPCRs. Examples are shown for cannabinoid
   receptor 1 (CB1) and metabotropic glutamate receptor 2 of which an
   array of known interactions with other GPCRs, ion channels, and
   classical signaling partners was identified (fig. S4A). Among the PPIs
   inferred from our coexpression analysis, we noticed a cluster of
   membrane-associated periodic skeleton (MPS) components (spectrin and
   ankyrin) and signaling molecules RTK (receptor tyrosine kinase) TrkB
   (tropomyosin-related tyrosine kinase B) and kinases Src (neuronal
   proto-oncogene tyrosine-protein kinase Src) and Fyn (tyrosine-protein
   kinase Fyn)] to pair with specific GPCRs, such as CB1 (fig. S4, A and
   B). This finding, in general, agreed with the recently reported
   colocalization of MPS components with CB1 and RTKs in neurons to form a
   cytoskeleton-dependent signaling platform ([126]29).

   To infer the unknown GPCR interaction network in the brain, we filtered
   our coexpression data to retain potential GPCR interactions with
   transmembrane proteins at a higher stringency (PCC, >0.9). This
   resulted in a high-quality predicted GPCR interaction network entailing
   2828 unique PPIs between 120 GPCRs and 1159 transmembrane proteins
   ([127]Fig. 4A and table S6). For CB1, we picked up an uncharacterized
   and most strongly correlated partner, leucine-rich repeat transmembrane
   protein FLRT3, for validation. Immunostaining of CB1 and FLRT3 in the
   hippocampal neuron revealed the colocalization of FLRT3 with a fraction
   of CB1-positive fibers ([128]Fig. 4B). Partial colocalizaiton of CB1
   and FLRT3 was also observed in the brain slice of HIP ([129]Fig. 4C).
   To further verify the interaction between CB1 and FLRT3, we carried out
   a proximity ligation assay (PLA) ([130]30) to confirm the close
   physical distribution between the two proteins in both CB1-expressing
   Chinese hamster ovary (CHO) cells and primary neurons (fig. S4C). In
   addition, immunoprecipitation (IP) was performed to verify the in vitro
   interaction between CB1 and FLRT3 (fig. S4D). These validation results
   clearly demonstrate the capacity of our proteomics-derived coexpression
   analysis to reveal new GPCR interactions that occur endogenously in the
   brain.

Fig. 4. A predicted GPCR interaction network from multiregional protein
coexpression analysis.

   Fig. 4
   [131]Open in a new tab

   (A) A predicted interaction network between 120 GPCRs and 1159
   transmembrane proteins in the mouse brain from high-stringency
   coexpression analysis (PCC >0.9). An enlarged PPI module for CB1 is
   shown, with the correlation coefficient for each putative PPI
   annotated. The most strongly correlated partner (FLRT3) was selected
   for experimental validation. (B) Colocalization of FLRT3 (red) with a
   fraction of CB1 (green) in the rat hippocampal neuron. Both endogenous
   proteins were stained with their antibodies. (C) Confocal image of
   FLRT3 (red) and CB1 (green) immunocytostaining in the hippocampal brain
   slice. Robust CB1-positive fibers were observed; a fraction of
   CB1-positive puncta colocalized with FLRT3 signals, which also showed a
   puncta staining pattern. Zoomed-in images of the boxed regions in (B)
   and (C) are shown on the right.

Profiling the brain region–resolved transmembrane proteome in a mouse
depression model

   To further demonstrate the power of our transmembrane proteome
   profiling technology for neurobiology, we applied our established
   workflow to a well-established mouse depression model to generate new
   insights into the pathophysiology of depression and accelerate the
   discovery of potential drug targets. Here, we adopted the chronic
   unpredictable mild stress (CUMS) model, in which CUMS mice were exposed
   chronically to a battery of unpredictable stressors and developed
   depressive-like symptoms after 21 days of consecutive exposure
   ([132]Fig. 5A). Specifically, CUMS mice exhibited reduced sucrose
   preference and increased immobility in behavioral tests, which
   indicated the depressive-like anhedonia [as measured by the sucrose
   preference test (SPT)] and despair [as measured by the tail suspension
   test (TST) and forced swimming test (FST)] (fig. S5).

Fig. 5. Transmembrane proteome profiling of the mouse brain in the CUMS model
of depression.

   Fig. 5
   [133]Open in a new tab

   (A) Procedure for establishing the CUMS model. (B) Anatomically
   dissected brain regions of the control and CUMS mice for proteomic
   analysis. (C) Number of protein IDs in the expanded DDA library and the
   GPCR hybrid library specifically built for the CUMS model. (D)
   Comparison of GPCR IDs from 11 brain regions of the CUMS model with the
   two libraries. (E) PCA (top) and HCA (bottom) of regional expression of
   all detected transmembrane proteins revealed tight clustering of
   control and CUMS groups for most of brain regions analyzed. Notice the
   separation of control and CUMS groups for PFC and HY in the PCA plot.
   (F) Number of up- and down-regulated GPCRs, ion channels, and
   transporters identified in each region (left). Number of unique DE
   proteins from the GPCR/ion channel/transporter families identified in
   total by multiregional analysis and by whole brain analysis (right).
   (G) Significantly enriched pathways (P < 0.001) in the DE GPCRs, ion
   channels, and transporters identified by multiregional analysis. cGMP,
   guanosine 3′,5′-monophosphate; PKG, cGMP-dependent protein kinase.

   For both the CUMS and control mice, we collected 11 anatomically
   dissected brain regions in triplicate, including the previously
   examined nine regions and two new regions [prefrontal cortex (PFC) and
   hypothalamus (HY)] ([134]Fig. 5B). Cell membrane fractionation, protein
   extraction, and digestion were performed using the same protocol, and
   each protein digest was analyzed by single-shot DIA MS ([135]Fig. 1B).
   For MS data mining, we constructed a largest DDA library specific for
   the CUMS model and a GPCR hybrid library derived from the DIA MS data,
   which contains the full complement of 524 mouse GPCRs ([136]Fig. 5C)
   (see Materials and Methods for details).

   After data filtering to control the subgroup FDR, an average of 74 GPCR
   proteins were identified and quantified per region with the GPCR hybrid
   library, representing an average gain of 25 GPCRs relative to the DDA
   library (fig. S6A). In total, 158 unique GPCRs were profiled in at
   least one brain region from control or CUMS mice, including 66 GPCRs
   exclusively detected with the targeted hybrid library strategy
   ([137]Fig. 5D). In addition, our study enabled in-depth profiling of
   180 ion channels and 179 transporters in the brain regions from control
   or CUMS mice (fig. S6B). Superior protein quantification consistency
   between experimental replicates of each brain region was achieved for
   both control and CUMS groups (median CV, 3.78 to 10.08%) (fig. S6C).

ID of differentially expressed transmembrane proteins in the depression model

   Both the PCA and unsupervised hierarchical clustering analysis (HCA)
   based on the quantification of 1998 transmembrane proteins revealed
   tight clustering of six replicates of the control and CUMS groups for
   most of the brain regions analyzed ([138]Fig. 5E). It implies that the
   molecular architecture of the transmembrane proteome was retained in
   most brain regions between the control and depressed states. We noticed
   the separation of control and CUMS groups for PFC and HY in the PCA
   plot, indicating larger perturbation of the transmembrane proteome here
   than in the other regions ([139]Fig. 5E, top).

   Our study identified discrete sets of differentially expressed (DE)
   transmembrane proteins from the GPCR, ion channel, and transporter
   families in each brain region of the CUMS model ([140]Fig. 5F and table
   S7). In accordance with the PCA plot, the largest number of DE
   transmembrane proteins were found in PFC and HY, two brain regions
   critical for mood regulation and stress response ([141]31–[142]34). A
   total of 91 up-regulated and 142 down-regulated unique proteins from
   the three families were identified from our brain proteome profiling of
   the CUMS model, which uncovered the most comprehensive landscape of
   transmembrane proteome remodeling associated with depression
   pathogenesis. In sharp contrast, only 16 dysregulated proteins were
   identified when analyzing the whole mouse brain of the CUMS model
   ([143]Fig. 5F and table S7). Therefore, the localized transmembrane
   proteome remodeling can be only captured by anatomical dissection
   combined with our high-sensitivity quantitative proteomics.

   This group of 233 DE transmembrane proteins identified in our study is
   enriched in 13 pathways, including synaptic vesicle cycle, nicotine
   addiction, oxidative phosphorylation, calcium signaling, and adenosine
   3′,5′-monophosphate (cAMP) signaling ([144]Fig. 5G). In addition, they
   are enriched in molecules known to be involved in the development of
   Parkinson’s disease and Alzheimer’s disease, which may indicate shared
   molecular mechanisms between neurodegeneration and depression, as noted
   previously ([145]35).

Discovery of two GPCR proteins as novel regulators of depression

   Given our particular interest in mining the GPCR family in search of
   the molecular regulators, we examined 35 GPCRs showing significant
   differential expression in PFC, HIP, or HY, three regions engaged in
   mood disorder and development of depression ([146]31–[147]34). After an
   extensive literature search, we were excited to find 19 DE GPCRs
   identified from the three regions turned out to be disclosed regulators
   of depressive-like behaviors and serve as potential antidepression
   targets, all uncovered by pharmacological intervention and/or genetic
   manipulation in vivo ([148]Fig. 6; individual target references in