Abstract

   Temporally regulated alternative splicing choices are vital for proper
   development, yet the wrong splice choice may be detrimental. Here, we
   highlight a previously unidentified role for the neurotrophin receptor
   splice variant TrkB.T1 in neurodevelopment, embryogenesis,
   transformation, and oncogenesis across multiple tumor types in humans
   and mice. TrkB.T1 is the predominant NTRK2 isoform across embryonic
   organogenesis, and forced overexpression of this embryonic pattern
   causes multiple solid and nonsolid tumors in mice in the context of
   tumor suppressor loss. TrkB.T1 also emerges as the predominant NTRK
   isoform expressed in a wide range of adult and pediatric tumors,
   including those harboring tropomyosin receptor kinase fusions. Affinity
   purification–mass spectrometry proteomic analysis reveals distinct
   interactors with known developmental and oncogenic signaling pathways
   such as Wnt, transforming growth factor–β, Sonic Hedgehog, and Ras.
   From alterations in splicing factors to changes in gene expression, the
   discovery of isoform specific oncogenes with embryonic ancestry has the
   potential to shape the way we think about developmental systems and
   oncology.
     __________________________________________________________________

   TrkB.T1 is oncogenic when overexpressed postnatally in many cell types
   that once expressed it during normal embryonic development.

INTRODUCTION

   Embryogenesis and neurodevelopment, in particular, comprise an elegant
   and well-orchestrated series of tightly regulated events, culminating
   in an organized and highly functioning organism. Cancer, on the other
   hand, can often be viewed as an uncontrolled, unrestrained, genetically
   chaotic disease, lacking the precise spatial and temporal rigidity
   associated with normal development. While appearing to be on opposite
   sides of the organizational spectrum, the similarities between early
   development and oncogenesis are numerous. This has been observed
   frequently as key developmental signaling pathways, such as Wnt,
   Hedgehog, or Notch, have been shown to be dysregulated in cancer across
   all stages from tumor initiation and maintenance to metastasis ([58]1).

   Encoded by the NTRK2 gene, the tropomyosin receptor kinase TrkB, which
   has many alternatively spliced isoforms, has well-established roles
   across neurodevelopment and astrocyte biology and has also been
   implicated in a wide range of cancer types ([59]2–[60]23). NTRK2 gene
   fusions have recently been the focus of many clinical and
   pharmacological studies ([61]24–[62]27), while inhibition of the TrkB
   kinase has been shown to inhibit cell proliferation and contribute to
   apoptosis ([63]7). In addition to well-known roles in neurobiology,
   recent studies suggest that NTRK2 may render cells resistant to anoikis
   and prone to metastasis, while others suggest its involvement in the
   epithelial-to-mesenchymal transition associated with increased
   migration and invasiveness of many cancer cell lines ([64]6, [65]8,
   [66]9, [67]28–[68]32), yet the precise mechanisms of how TrkB exerts
   its oncogenic role are not fully known. While prior studies have been
   fundamental in uncovering NTRK2 involvement in cancer, the complex
   posttranslational modifications intricate splicing patterns, and prior
   roles in embryogenesis are often ignored.

   Recent findings have shown that a kinase-deficient NTRK2 splice
   variant, TrkB.T1, is the predominant neurotrophin receptor in human
   gliomas, enhances glioma aggressiveness in mice, and increases the
   perdurance of platelet-derived growth factor (PDGF)–induced Akt and
   signal transducers and activators of transcription 3 (STAT3) signaling
   ([69]33). This splice variant contains a unique terminal exon encoding
   a 11–amino acid tail that is 100% conserved across species including
   chicks, mice, rats, felines, and humans ([70]34–[71]39), suggesting a
   potential evolutionarily important biological role. In light of recent
   findings implicating this kinase-deficient splice variant in glioma
   biology ([72]33), we explore the role of TrkB.T1 in neural development
   and early embryonic central nervous system (CNS) and mesenchymal
   development in particular.

   In this study, we characterize the role of the NTRK2 splice variant,
   TrkB.T1, in embryonic neurodevelopment and early organogenesis via
   transcript analysis of an enhanced single-cell combinatorial indexing
   RNA sequencing analysis method (sci-RNA-seq3) data of embryonic day 9.5
   (E9.5) to E13.5 mouse embryos ([73]40) and transcript-specific
   immunohistochemistry. We demonstrate a developmental role for TrkB.T1
   beyond that of the CNS, highlight TrkB.T1’s transformation potential in
   vitro, and uncover its role as an oncogenic driver in vivo using a
   RCAS-TVA mouse model ([74]41, [75]42). We further reveal TrkB.T1 to be
   the predominant TRK isoform expressed across the majority of adult and
   pediatric tumors using clinical data from The Cancer Genome Atlas
   (TCGA) ([76]43) and the Therapeutically Applicable Research to Generate
   Effective Treatments (TARGET) initiative, respectively, and uncover a
   several potential interactors via proteomic analysis. Using TrkB.T1, we
   explore the role of developmentally regulated splicing choices and
   subsequently show that forced expression of a specific isoform can be
   neoplastic in the same postnatal organs that once expressed it during
   embryogenesis. This kinase-deficient splice variant of NTRK2 is found
   across normal embryonic development and is a driver of cancer in vivo
   when overexpressed postnatally in many of the cell types that expressed
   it during embryonic development.

RESULTS

TrkB.T1 in embryonic development and organogenesis

   It is widely known that the majority of organ development in rodents
   occurs before E18 ([77]40) and recent advances in next-generation
   sequencing have allowed for the characterization of transcriptional
   dynamics across development at a single-cell resolution resulting in
   the Mouse Organogenesis Cell Atlas (MOCA) ([78]40). To explore the
   expression of TrkB.T1 in CNS development, we used the fact that the
   TrkB.T1 and full-length TrkB.FL NTRK2 transcripts differ in their 3′
   ends ([79]38), which allows for transcript quantification in
   single-cell RNA sequencing data and mapped expression of individual
   transcripts in mouse embryos over developmental time. Specifically,
   TrkB.T1 is not the result a pure truncation and contains a uniquely
   spliced in exon that is not present in the full-length,
   kinase-containing variant ([80]Fig. 1A). TrkB.FL contains several exons
   downstream of this exon that are not present in TrkB.T1 ([81]Fig. 1A),
   allowing for quantification of these two alternative spliced variants
   in our dataset ([82]38). Using sci-RNA-seq3 data, the MOCA ([83]40)
   provides gene expression data and cell trajectory annotation for
   2,026,641 cells from 61 mouse embryos across five development stages
   (E9.5 to E13.5), resulting in 38 main cell clusters and 10 main
   trajectories. The expression of transcripts for all protein coding
   genes, including TrkB.T1 and TrkB.FL, was normalized and visualized
   over the existing landscape provided by Cao et al. ([84]40). First, we
   examined TrkB.FL expression patterns in the t-distributed stochastic
   neighbor embedding (t-SNE) plot of all 38 cell types within the MOCA
   and found that TrkB.FL expression is appreciably restricted to a subset
   of cell types, all of which were neuronal in nature ([85]Fig. 1, B and
   C, and fig. S1, A to E). TrkB.T1, however, was expressed broadly across
   many different cell types within the embryo during this period
   ([86]Fig. 1, D and E, and fig. S1, F to J). We next sought to explore
   the NTRK2 transcript expression patters within the CNS developmental
   trajectory and found that TrkB.T1 cellular expression is observed
   across multiple neurodevelopmentally driven cell clusters, including
   radial glia, immature oligodendrocytes, and isthmic organizer cells
   ([87]Fig. 1F). The second largest population of cells with high
   variable levels of TrkB.T1 expression during this developmental period
   is found in the mesenchymal trajectory, and mapping expression of both
   transcripts showed a vast predominance of TrkB.T1 over TrkB.FL similar
   to the nonneuronal cell types of the developing CNS ([88]Fig. 1, F to
   I). Visualization of NTRK2 expression reveals intense, diffuse patterns
   of TrkB.T1 across multiple days in multiple cell types, with expression
   levels rivaling those of actin in particular clusters (figs. S1, F to
   J, and S2), whereas TrkB.FL expression remains predominantly restricted
   to mature neurons (fig. S1, A to E). An interactive site allowing
   three-dimensional visualization and further exploration of NTRK2
   transcripts among cell types and within the CNS and mesenchymal
   trajectories over developmental time can be found at
   [89]https://atlas.fredhutch.org/fredhutch/ntrk2/.

Fig. 1. sci-RNA-seq3 transcript analysis across E9.5 to E13.5 mouse embryonic
development shows TrkB.FL expression in neuronal clusters with widespread
TrkB.T1 expression in CNS and mesenchymal trajectories.

   [90]Fig. 1.
   [91]Open in a new tab

   (A) Schematic depicting two alternatively spliced NTRK2 transcripts
   (TrkB.T1 and TrkB.FL) shows unique exons present in each variant. t-SNE
   visualization of TrkB.FL expression (B and C) and TrkB.T1 (D and E) of
   2,026,641 mouse embryo cells from E9.5 to E13.5 ([92]40) shows high
   TrkB.FL expression in neuronal clusters and CNS clusters with strong,
   diffuse TrkB.T1 expression across the majority of cell clusters.
   Visualization of CNS (F) and mesenchymal (G) cell type trajectories
   across all days shows strong expression of TrkB.FL in mature neuronal
   cells (H) with diffuse expression of TrkB.T1 in large subsets of
   immature stem-like cells across developmental trajectories (I).

   As the sci-RNA-seq3 data show widespread TrkB.T1 expression across
   multiple clusters within and outside the CNS, we wanted to visualize
   TrkB.T1 and TrkB.FL expression across embryonic development beyond
   E13.5. To this end, we stained histological sections of embryos from
   E10 to E17 with antibodies specific to the TrkB.FL kinase domain and an
   antibody specific to the intracellular 11–amino acid tail of TrkB.T1
   ([93]44). Similar to the sci-RNA-seq3 data, examination at early
   embryonic time points, when the most rapid cellular growth and
   organogenesis are underway, the protein ratio between these two
   receptors is different, with TrkB.T1 present at exceedingly high levels
   in multiple organ sites ([94]Fig. 2, A to I) compared to TrkB.FL
   ([95]Fig. 2, J to R), which is predominantly expressed and maintained
   in CNS only, consistent with the transcript levels for the two TrkB
   gene products shown in [96]Fig. 1. By the end of embryonic development
   circa E17, immunohistochemistry of both splice variants reveals that
   TrkB.T1 levels have decreased but are still present throughout various
   non-CNS organs, while the majority of TrkB.FL remains restricted within
   the CNS ([97]Fig. 2). These data suggest that in addition to known
   roles in CNS development and gliomagenesis, TrkB.T1 may have a more
   widespread role in early organogenesis throughout the developing
   organism. Since TrkB.T1 has been shown to play a role brain tumors
   ([98]33), this widespread embryonic expression pattern suggests that
   TrkB.T1’s roles, in both development and cancer, may not be restricted
   to the CNS, highlighting the possibility that it may also influence
   additional tumor types beyond what has been shown for gliomas.

Fig. 2. TrkB transcript expression across E10 to E17 embryonic development.

   [99]Fig. 2.
   [100]Open in a new tab

   Representative micrographs of immunohistochemical staining of TrkB.T1
   in the whole-mount murine embryo section at E11 (A) and E16 (B).
   Quantitative assessment of staining intensities in CNS (red), muscle
   (green), and liver (blue) at different developmental stages (C).
   Representative micrographs for TrkB.T1 staining of the (D) brain, (E)
   lung, (F) kidney, (G) liver, (H) small intestine, (I) skeletal muscle
   at E16. Immunohistochemical staining of TrkB.FL in the whole-mount
   embryo section at E11 (J) and E16 (K) with (L) quantitative assessment
   of staining intensities in the CNS (red), muscle (green), and liver
   (blue) at different developmental stages. Representative micrographs
   for TrkB.FL staining in the (M) brain, (N) lung, (O) kidney, (P) liver,
   (Q) small intestine, (R) skeletal muscle at E16. Scale bars, 1 mm (A,
   B, J, and K) and 100 μm (D to I and M to R).

TrkB.T1 exerts significantly more transformation capacity compared to TrkB.FL
in colony formation assay and is required for neurosphere formation

   Given the widespread distribution of TrkB.T1 in multiple organ sites
   and cell types across embryonic development and our recent findings
   that the TrkB.T1 splice variant predominates over the full-length
   kinase-containing TrkB variant in gliomas and that this TrkB.T1 gene
   product enhances PDGF-driven glioma aggressiveness in vivo and
   amplifies PDGF-driven AKT and Stat signaling in vitro ([101]44), we
   wanted to compare the relative transformation potential of the
   kinase-containing TrkB.FL and kinase-deficient TrkB.T1 isoforms in a
   gain-of-function experiment using a soft agar colony formation assay.
   Forced expression of the TrkB.T1 splice variant containing its unique
   11–amino acid intracellular tail in 3T3 cells led to significantly
   enhanced colony formation (average colonies = 209) compared to
   overexpression of the kinase-containing variant, TrkB.FL (average
   colonies = 48), or control vector (average colonies = 19.33) [F(2,6) =
   401.5, P < 0.0001) ([102]Fig. 3A). Using a short hairpin designed
   against TrkB.T1 to explore loss of function, we show that TrkB.T1 is
   necessary for neurosphere formation of primary N/tv-a;Ink4a/Arf^−/−
   neural stem cells (NSCs) and that knocking down TrkB.T1 with
   RCAS-PDGFB-shTrkB.T1 in primary NSCs leads to significantly less
   neurosphere formation compared to RCAS–PDGFB–shScrambled (SCR) control
   ([103]Fig. 3B and fig. S3).

Fig. 3. Gain- and loss-of-function experiments show that TrkB.T1 transforms
3T3 cells, is required for neurosphere formation, and causes solid and
nonsolid tumors throughout the body in mice when overexpressed in the context
of tumor suppressor loss.

   [104]Fig. 3.
   [105]Open in a new tab

   (A) Soft agar colony formation assay in 3T3 cells shows significantly
   increased colonies with TrkB.T1 compared to control and TrkB.FL (P <
   0.0001; graph represents means + SEM). (B) TrkB.T1 knockdown with
   RCAS-PDGFB-shTrkB.T1 significantly decreases neurosphere formation in
   primary Nestin/tv-a;Ink4a/Arf^−/− NSCs (RCAS-PDGFB-shTrkB.T1: M =
   16.833 + SD 2.317 versus RCAS-PDGFB-shSCR: M = 5.000 + SD 2.280;
   t(5)=11.31, P < 0.0001). Survival curves for Nestin/tv-a;Ink4a/Arf^−/−
   mice injected with RCAS-shPTEN alone or RCAS-shPTEN + RCAS-TrkB.T1 (C)
   demonstrate decreases in survival due to tumor burden outside the CNS
   for mice injected with RCAS-TrkB.T1 + RCAS-shPTEN versus RCAS-shPTEN
   alone (median survival, 131 versus 250 days; log-rank hazard ratio,
   0.07243; 95% confidence interval, 0.0.04768 to 0.11; P < 0.0001)
   including the soft tissue, kidney, liver, and spleen as shown in (D)
   and (E). Scale bars, 100 μm (100×) and 20 μm (600×). Flow cytometry
   analysis of splenic tumor demonstrates a large B cell lymphoma [(CD45
   bright/B220^+; blue population) with diffuse collection of large
   atypical cells with vesicular chromatin and high mitotic rate with a
   vague nodular pattern], myeloid leukemia [(CD45^+/GR1^+; pink
   population) with extramedullary erythropoiesis (Ter119^+/dim CD45; gray
   population), showing focal collections of intermediate size cells with
   open chromatin and frequent mitosis, admixed with maturing erythroid
   cells], and an abnormal T cell population [(CD45
   bright/CD3e^+/CD4^−/CD8a^−; green population) consistent with a T cell
   lymphoma admixed with reactive B cells (CD45 bright/B220 variable; blue
   population), showing large atypical cells with a high mitotic rate
   admixed with scattered small reactive lymphocytes and eosinophils]
   (20×). Scale bars, 50 μm. ***P < 0.0001.

Forced postnatal expression of TrkB.T1 causes multiple cancer types in mice

   Prior investigation into TrkB.T1’s role in gliomas revealed that a
   multitude of genes in phosphatidylinositol 3-kinase
   (PI3K)/Akt/phosphatidylinositol phosphate/inositol phosphate pathways
   were significantly positively correlated with NTRK2 expression in low
   grade glioma (LGG) and glioblastoma multiforme (GBM) compared to normal
   brain ([106]44) via differential gene correlation analysis (DGCA),
   suggesting a role for TrkB.T1 in the PI3K signaling pathway. On the
   basis of these results, we wanted to explore whether TrkB.T1 is capable
   of inducing tumors within and outside the CNS in the context of PTEN
   loss, because Pten (phosphatase and tensin homolog) is a negative
   regulator of Akt. We chose mice with a sensitized background [CDKN2A
   (Ink4a/Arf) loss, as this is a frequent event in multiple cancers]
   because TrkB.T1 overexpression does not cause tumors on its own in
   wild-type mice ([107]33). To answer this question of whether TrkB.T1
   causes tumors outside the CNS, we used the RCAS/TVA system, which
   allows for somatic expression of a gene of interest in particular cell
   types, to force overexpression TrkB.T1 and knockdown PTEN expression in
   nestin-positive progenitor cells of Ink4a/Arf^−/− mice ([108]41).

   Misexpression of factors or proteins that are benign or necessary in
   one cell type may lead to aberrant and noncanonical signaling in
   another cell type, as has been shown for Wnt signaling ([109]45).
   Because nestin is expressed in a wide range of stem and progenitor
   cells during the development of various organ sites (not just neural
   and glial progenitors), this promoter was chosen to target RCAS
   delivery specifically to nestin-positive cells. N/tv-a;Ink4a/Arf^−/−
   mice were injected intraperitoneally or intramuscularly with
   RCAS-TrkB.T1 and RCAS-shPTEN. While neither RCAS-shPTEN nor
   RCAS-TrkB.T1 alone caused cancer in this genetic background during the
   time course of this experiment, the combination of RCAS-shPTEN and
   RCAS-TrkB.T1 injections caused a range of solid and nonsolid tumors
   originating from nestin-positive cells outside of CNS ([110]Fig. 3C).
   These included soft tissue sarcomas, carcinomas arising in the kidney,
   and lymphoid leukemias and lymphomas ([111]Fig. 3, D and E). Because
   our RCAS-TrkB.T1 + RCAS-shPTEN–injected mice exhibited a wide range of
   tumor types (sarcomas, kidney, liver, nonsolid tumors, etc.), we
   injected many more mice in this cohort than was needed for
   statistically significant survival differences so that we could
   ultimately have a wide range of tumor types for subsequent histological
   analyses. Immunohistochemistry on these solid tumors confirmed that
   these tumors throughout the body had both high levels of TrkB.T1 and
   loss of PTEN expression (fig. S4, A to D), demonstrating that forced
   expression of TrkB.T1 in combination with PTEN loss has the potential
   to form tumors in the same organ sites that once expressed TrkB.T1
   embryonically.

Proteomic analysis reveals TrkB.T1-specific interactors with known
developmental, oncogenic, and cell cycle signaling pathways

   We explored potential interactors and up-regulated pathways in an
   isoform-specific NTRK2 manner in 3T3 cells separately transduced with
   pLJM1-lentiviral vectors containing either TrkB.T1–hemagglutinin (HA)
   or TrkB.FL-HA ([112]46, [113]47) in the presence or absence of serum.
   Lysates were collected and subjected to HA-tag pull-down and subsequent
   affinity purification–mass spectrometry (AP-MS) proteomic analysis
   ([114]48) to reveal potentially directly or indirectly interacting
   proteins ([115]48).

   Using HA-tagged NTRK2 splice variants as bait, all immunoprecipitated
   proteins were identified, quantified, characterized, and subjected to
   Gene Ontology (GO) and Reactome pathway enrichment analysis to
   understand and explore potential protein-protein interactions
   ([116]Fig. 4). In the absence of serum, TrkB.FL-expressing 3T3 cells
   began to die off, while TrkB.T1 cells continued to thrive, similar to
   the enhanced growth observed in the colony formation assay ([117]Fig.
   3A). For these serum-starved conditions, 163 shared proteins were
   pulled down in complexes with both TrkB.T1 and TrkB.FL, 110 were pulled
   down specifically with TrkB.FL, and 720 proteins were pulled down
   specifically in complexes with TrkB.T1. Under normal culture
   conditions, in the presence of serum, 749 proteins were pulled down in
   complexes with both TrkB.T1 and TrkB.FL, 399 were pulled down
   specifically with TrkB.FL, and 565 proteins were pulled down
   specifically in complexes with TrkB.T1 (dataset S1). GO and Reactome
   pathway enrichment analyses on these clusters of affinity-purified
   proteins revealed differential expression of a host of developmental,
   oncogenic, and mitotic pathway terms associated specifically with
   TrkB.T1 in the presence or absence of serum compared to the terms
   associated with TrkB.FL (fig. S5 and dataset S1). Shared developmental
   and oncogenic terms between TrkB.T1 and TrkB.FL in the presence of
   serum implicated proteins or classes of proteins involved in fibroblast
   growth factor receptor 2 (FGFR2), insulin-like growth factor receptor
   (IGFR), SLITs, Roundabout (ROBO), Sonic Hedgehog (SHH), tumor protein
   53 (TP53), and runt-related transcription factor (RUNX) signaling, long
   known to be associated with neurotrophin signaling. Additional classes
   of proteins were found to be TrkB.T1 specific under normal culture
   conditions and maintained in the absence of serum. Among these
   TrkB.T1-specific interactors (n = 565) are proteins with known
   developmental and oncogenic roles, including Gli, Hedgehog, Wnt,
   transforming growth factor–β (TGF-β), Ras, mitogen-activated protein
   kinase (MAPK), MET, calmodulin (CaM), and other signaling pathways
   ([118]Fig. 4, fig. S5, and dataset S1). The distinct protein sets were
   pulled down with TrkB.T1 highlight a host of developmental, oncogenic,
   and cell cycle/cell division pathway binding partners that were not
   previously known to interact with TrkB.T1 along with several known
   interactors such as those associated with RhoA ([119]49, [120]50) and
   Ca^2+ signaling ([121]51).

Fig. 4. AP-MS spectrometry reveals NTRK2 splice variant interactors
associated with developmental and oncogenic pathways.

   [122]Fig. 4.
   [123]Open in a new tab

   3T3 cells were transduced with TrkB.T1-HA or TrkB.FL-HA, each in
   triplicate, and subjected to HA pull-down to identify direct and
   indirect interactors via AP-MS as shown in fig. S3 and dataset S1.
   Proteins specific to each isoform and shared between both isoforms
   (with serum: n = 599 for TrkB.T1, n = 399 for TrkB.FL, and n = 749
   shared; without serum: n = 720 for TrkB.T1, n = 110 for TrkB.FL, and n
   = 163 shared) were subjected to GO and Reactome pathway enrichment
   analysis to find proteins associated with developmental, oncogenic, and
   cell cycle/cell division pathways. *Developmental, oncogenic, and cell
   cycle/cell division–associated signaling pathways listed are associated
   under normal culture conditions (with serum) unless shown with
   asterisk, which denotes that proteins associated with these pathways
   were found both under normal culture conditions and serum-starved
   conditions.

TrkB.T1 is the predominantly expressed form of TrkB RNA and protein in adult
and pediatric tumors

   The large number and breadth of murine tumor types induced by TrkB.T1
   expression, combined with the knowledge that, in gliomas, the
   predominant form of TrkB is the TrkB.T1 splice variant, prompted an
   in-depth pan-cancer analysis of NTRK2 transcript expression across all
   organ sites within TCGA. Similar to what has been reported for gliomas
   ([124]44), expression of the full-length, kinase-containing TrkB.FL is
   minimal to absent across the majority of tumor types within TCGA, while
   TrkB.T1 expression is consistently high ([125]Fig. 5 and dataset S2).
   In-depth transcript characterization of all TCGA tumors harboring known
   NTRK fusions recapitulates this pattern and shows that, regardless of
   TRK fusion (NTRK1, NTRK2, and NTRK3), expression of the
   TrkB.T1-specifiic exon (encoding for the unique 11–amino acid
   intracellular region) remains consistently higher than all
   kinase-containing TRK exons, regardless of the particular TRK fusion or
   tumor type (fig. S6 and dataset S3).

Fig. 5. Pan-cancer NTRK2 transcript analyses.

   [126]Fig. 5.
   [127]Open in a new tab

   NTRK2 transcript analysis shows minimal to zero expression of TrkB.FL
   transcript and increased expression for TrkB.T1 transcript across all
   TCGA organ sites: adrenocortical carcinoma (ACC), bladder urothelial
   cancer (BLCA), breast invasive carcinoma (BRCA), cervical squamous cell
   carcinoma and endocervical adenocarcinoma (CESC), cholangiocarcinoma
   (CHOL), colon adenocarcinoma (COAD), colorectal adenocarcinoma
   (COAD/READ), lymphoid neoplasm diffuse B cell lymphoma (DLBC),
   esophageal carcinoma (ESCA), head and neck squamous carcinoma (HNSC),
   kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC),
   kidney renal papillary cell carcinoma (KIRP), acute myeloid leukemia
   (LAML), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma
   (LUAD), lung squamous cell carcinoma (LUSC), mesothelioma (MES),
   ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma
   (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate
   adenocarcinoma (PRAD), rectum adenocarcinoma (READ), sarcoma (SARC),
   skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), stomach
   and esophageal (STES), testicular germ cell tumor (TGCT), thyroid
   carcinoma (THCA), thymoma (THYM), uterine corpus endometrial carcinoma
   (UCEC), uterine carcinoma (UCS), and uveal melanoma (UVM). Data are
   represented as boxplots where the middle line is the median, the lower
   and upper hinges correspond to the first and third quartiles (the 25th
   and 75th percentiles), the upper whisker extends from the hinge to the
   largest value no further than 1.5× IQR from the hinge (where IQR is the
   interquartile range or distance between the first and third quartiles),
   and the lower whisker extends from the hinge to the smallest value at
   most 1.5× IQR of the hinge, while data beyond the end of the whiskers
   are outlying points that are plotted individually.

   To extend upon these transcript results demonstrating predominant
   TrkB.T1 expression across a wide range of human tumors, we next
   determined how well RNA transcripts correlate with TrkB.T1 protein
   distribution. Immunohistochemistry on a tissue microarray of various
   adult tumors ([128]Fig. 6) reveals variable, increased TrkB.T1
   expression and distribution, with high TrkB.T1 H score values (dataset
   S4) for nearly all tumor types. These results show that both RNA
   transcript expression and protein levels for TrkB.T1 are high across
   all tumor types in available datasets, suggesting that this is the
   NTRK2 splicing choice made across cancers and predominates in
   nonneuronal cells of the embryo as well.

Fig. 6. TrkB.T1 has variable expression across multiple human tumor types.

   [129]Fig. 6.
   [130]Open in a new tab

   Representative micrographs of TrkB.T1 expression in (A) clear cell
   renal cell carcinoma, (B) colonic adenocarcinoma, (C) leiomyosarcoma,
   (D) mammary carcinoma (ductal carcinoma in situ is shown here), (E)
   pancreatic adenocarcinoma, and (F) seminoma. Scale bars, 100 μm.

   In addition to immunohistochemistry ([131]Fig. 6) and the data from
   TCGA, which is composed of mostly adult tumors, previous studies have
   also shown a potential role for TrkB in pediatric tumors, and its gene
   products have been shown to be expressed at high levels in high-risk
   neuroblastoma (NBL) and correlated with poor prognosis ([132]29,
   [133]52), while levels of TrkB gene products have been shown to be
   correlated with unfavorable outcome in patients with Wilms tumor (WT)
   ([134]15). Transcript-specific analysis of pediatric data from the
   TARGET program confirms that, similar to adult tumors in TCGA, TrkB.T1
   expression is the predominant isoform compared to TrkB.FL in several
   cancer types, including WT, rhabdoid tumor, NBL, and clear-cell sarcoma
   of the kidney (fig. S7 and dataset S2).

   Together, these data show that TrkB.T1 is the NTRK2 isoform expressed
   in multiple cell types across embryonic development, is expressed at
   unexpectedly high rates across adult and pediatric cancers from
   multiple sites within TCGA and TARGET ([135]Fig. 5, fig. S7, and
   datasets S2 to S4), and is capable of transforming cells in vitro and
   causing tumors across various organ sites in the context of loss of the
   common tumor suppressors Ink4a/Arf and PTEN in mice ([136]Fig. 3 and
   fig. S4), potentially through its interactions with known oncogenic and
   developmental signaling pathways (fig. S5 and dataset S1). These data
   highlight the TrkB.T1 splice variant as one candidate whose aberrant
   expression with concurrent tumor suppressor loss causes cancer in the
   same developed organs where it played a role in early embryonic
   development.

DISCUSSION

   While many point mutations have been found in cancer, only a small
   majority of these mutations have been shown to be causal. These rare
   driver mutations contribute to oncogenic signaling by altering pathways
   that are otherwise tightly regulated in normal tissues ([137]53). It is
   known that cancer cells exhibit vast transcriptomic alterations and
   that cancer-specific isoforms are not merely by-products of abnormal
   physiology ([138]54), yet most cancer-associated splicing choices, such
   as point mutations, are not oncogenic drivers.

   Differential splicing choices allow for highly regulated expression of
   specific transcripts in normal development, yet expression of specific
   splice variants at developmentally inappropriate time points or in the
   wrong cell types may be detrimental. While it is known that aberrant
   splicing mechanisms exist in cancer ([139]55–[140]57), in this study,
   we show that a splice variant with a physiologic role in early
   organogenesis, but whose expression is not found in corresponding
   mature tissue, can be a driver of cancer. TrkB.T1 appears to be a
   candidate whose regulated splicing choice appears in normal embryonic
   development and whose misexpression at later postnatal time points is
   oncogenic.

   While the embryogenic data offer correlative insight, the causal
   RCAS-TVA mouse studies show that aberrant expression of TrkB.T1 causes
   neoplasms at multiple primary sites, potentially by trapping cells in
   an undifferentiated state similar to their early embryogenic stage via
   interactions with known signaling pathways such as Wnt, SHH, TGF-β, or
   Ras. In the embryonic sci-RNA-seq3 data, TrkB.FL and TrkB.T1 are
   rarely, if ever, expressed in the same cell, and they also have a
   unique set of interactors as evidenced by HA AP-MS, suggesting that
   TrkB.T1 does not function through transactivation of TrkB.FL but
   potentially through other signaling pathways. In the context of tumors
   in vivo, however, there are many endogenous receptor tyrosine kinases
   that could be influenced by the overexpression of TrkB.T1 during tumor
   initiation, maintenance, and metastasis. Future work should seek to
   address the precise interactors in vivo across a wide range of organ
   sites in a context- and cell-specific manner to expand upon initial
   experiments performed here in vitro with 3T3 cells and NSCs, to
   understand TrkB.T1’s complex influences on endogenous tumor cell
   proliferation and growth.

   These data highlight the TrkB.T1 splice variant as a promising
   candidate for future studies investigating the role of NTRK2 in
   development and oncogenesis across a range of organ sites and tumor
   models beyond the known neurobiological roles of TrkB.T1
   ([141]58–[142]60) and further strengthen the potential links between
   the tightly regulated splicing patterns of development with the
   aberrant splicing patterns observed in cancer. Approaches to studying
   splicing factors and alternatively spliced transcripts such as those
   delineated here should not be restricted to particular genes, stages of
   development, or organ sites and may uncover promising new avenues for
   diagnostics or therapeutics by revealing additional developmentally
   regulated, oncogenic splice variants.

METHODS

Bioinformatic analysis

Quantification of transcript data from sci-RNA-seq3

   SAM (Sequence Alignment/Map) alignment files for the MOCA dataset were
   downloaded from
   [143]https://shendure-web.gs.washington.edu/content/members/cao1025/pub
   lic/nobackup/. The “cell_annotation.csv,” which contained t-SNE
   coordinates, Uniform Manifold Approximation and Projection (UMAP)
   coordinates and information about clusters and trajectories, was
   downloaded from
   [144]https://oncoscape.v3.sttrcancer.org/atlas.gs.washington.edu.mouse.
   rna/downloads. The “Comprehensive gene annotation” file was downloaded
   from [145]www.gencodegenes.org/mouse/release_M12.html. For each of the
   2,062,641 cells, we calculated the number of strand-specific unique
   molecular identifies (UMIs) for each cell mapping to transcripts of
   each gene with the Python v.2.7.13 HTSeq ([146]61) package using
   GENCODE vM12 ([147]62). Monocle3 ([148]63) was used to construct a
   cell_data_set object with all the transcript expression data, and
   preprocess_cds() was used to normalize the transcript expression. The
   expression of TrkB.T1 and TrkB.FL was visualized over existing UMAPs
   and t-SNEs provided by Cao et al. ([149]40) using ggplot2 ([150]64). An
   interactive website ([151]https://atlas.fredhutch.org/fredhutch/ntrk2/)
   has been put together to facilitate further exploration of the NTRK2
   transcript data across various trajectories and cell types.

Obtaining and transforming transcript-level data from TCGA and TARGET

   Transcript data (TCGA RNAseqV2 RSEM data) for TCGA and TARGET organ
   sites were downloaded from Broad’s FireBrowse website
   ([152]https://gdac.broadinstitute.org/). Transcript data (TPM data)
   were also downloaded for TARGET from UCSC Xena
   ([153]https://xenabrowser.net/datapages/?dataset=target_RSEM_gene_tpm&h
   ost=https%3A%2F%2Ftoil.xenahubs.net&removeHub=https%3A%2F%2Fxena.treeho
   use.gi.ucsc.edu%3A443). The results published here are in whole or part
   based on data generated by TCGA ([154]43) and the TARGET
   ([155]https://ocg.cancer.gov/programs/target) initiative, phs000218.
   The data used for this analysis are available at
   [156]https://portal.gdc.cancer.gov/projects.

   To make the RSEM data from TCGA and TARGET and TPM data comparable, we
   converted RSEM counts from TCGA to TPM counts using the following
   formula ([157]65): RSEM can be multiplied by 10^6. For transcript
   analyses, NTRK2 transcript IDs were manually aligned to confirm
   sequence homology and are as follows: TrkB.FL (UCSC, uc004aoa.1;
   Ensembl, ENST00000376213.1, NTRK2_201) and TrkB.T1 (UCSC, uc004aob.1;
   Ensembl, ENST00000395882.1, NTRK2_204). Transcript data were visualized
   using boxplots using R package ggplot2 ([158]64). Gene fusions for all
   TCGA fusions were obtained from Genomic Data Commons portal
   ([159]https://portal.gdc.cancer.gov/). The transcript levels for
   TrkB.T1 and N-terminal truncated TrkB.T1, both containing the 11–amino
   acid intracellular region, for TCGA samples IDs containing TRK gene
   fusions (including fusions in NTRK1, NTRK2, and NTRK3) were compared to
   the mean of all other NTRK transcripts.

AP-MS analysis

   GO and Reactome pathway enrichment analysis for genes associated with
   enriched MS proteins ([160]66, [161]67) was done using R Bioconductor
   Packages clusterProfiler v 3.4.4 ([162]68), and dot plots were made
   using R Bioconductor package DOSE.

Generation of murine tumors

   The RCAS-TVA system used in this work has been described previously for
   murine tumor modeling in immunocompetent mice ([163]69). DF-1 cells
   were transfected with the relevant RCAS viral plasmids using
   Extreme-Gene 9Transfection reagent (Roche) accordingly to the
   manufacturer’s protocol. The cells were maintained for three passages,
   as described above to ensure viral propagation to all cells. After
   confirmation of RCAS-inserts by Western blot, DF-1 cells (passage 4 or
   later) were used for injection. Newborn N/tv-a;Ink4a/Arf^−/− pups
   [postnatal day 0 (P0) to P1] were injected (Hamilton syringe #84877)
   with 1 μl of approximately 1 × 10^5 DF-1 cells infected with and
   producing relevant RCAS viruses suspended in serum-free Dulbecco’s
   modified Eagle’s medium (DMEM; TrkB.T1 and shPTEN) ([164]41).
   Simultaneous delivery of two RCAS viruses was performed by the
   injection of 1 μl of approximately 2 × 10^5 DF-1 cells mixed with equal
   ratio. Mice were monitored for the duration of the study (250 days) to
   check for tumor-related symptoms such as palpable masses, lethargy,
   weight loss, seizure, hyperactivity, altered gait, poor grooming,
   macrocephaly, and paralysis. Mice with severe hydrocephalus presumably
   due to injection trauma or an inflammatory response against the DF-1
   cells were excluded from survival analysis in this study. All animal
   experiments were approved by and conducted in accordance with the
   Institutional Animal Care and Use Committee of Fred Hutchinson Cancer
   Research Center (protocol #50842).

Mouse tissue processing

   Mouse tissue (including normal tissue and solid tumors) was removed,
   fixed in 10% neutral-buffered formalin for a minimum of 72 hours, and
   embedded into paraffin blocks. Five-micrometer serial sections were cut
   from formalin-fixed paraffin-embedded specimens and mounted on slides.

Immunohistochemistry

   Immunohistochemical staining was performed on 5-μm
   formalin-fixed/paraffin-embedded tissue sections using a Discovery XT
   Ventana Automated Stainer (Ventana Medical Systems Inc.), run using
   standard Ventana reagents and Vector secondaries for staining with the
   TrkB.T1 SPEH1_D12 scFv-Fc fusion ([165]44) at 1:500 and PTEN (Cell
   Signaling Technology, #9188, lot no. 4) at 1:100 or TrkB (kinase
   specific against amino acid 810; Abcam, #ab18987, lot no. GR3280550-2)
   at 1:250. All embryonic histology slides and human tissue microarray
   (BioMax, US, #BC001134b) slides were scanned using a Ventana DP200
   slide scanning system (Roche Diagnostics) at ×20 magnification. Digital
   images of different organs during mouse development were analyzed using
   the Fiji image analysis software as described previously ([166]70,
   [167]71). Staining intensities in human tumor tissues were assessed
   using a semiquantitative H score system by multiplying the intensity of
   the stain (0, no staining; 1, weak staining; 2, moderate staining; 3,
   intense staining) by the percentage (0 to 100) of cells, showing
   staining intensity (H score range, 0 to 300) as described previously
   ([168]72).

Flow cytometry

   Rodent tumors were harvested and resuspended as single cells in
   phosphate-buffered saline (PBS)/0.3% bovine serum albumin. Cells were
   washed and incubated with the following antibodies: CD45-APC-Cy7
   (BioLegend, clone 30-F11, catalog no. 103116, lot no. B242535), CD3e-PE
   (BD, clone 145-2C11, catalog no. 553061, lot no. 22126), GR1-PerCP (BD,
   clone Ly-6G/Ly-6C, catalog no. 552093, lot no. 73108), TER119-APC (BD,
   clone Ter-119, catalog no. 557909, lot no. 42622), B220–Alexa Fluor 647
   (BD, clone RA3-6B2, catalog no. 557683, lot no. 22218), CD4-PerCP (BD,
   clone L3T3, catalog no. 553654, lot no. 60912), and CD8a-FITC (BD,
   clone Ly-2, catalog no. 553030, lot no. 46675), as indicated. Cells
   were analyzed on a custom-built LSR II flow cytometer (BD). Data
   compensation and analysis were performed using noncommercial software
   developed in our laboratory ([169]73).

Soft agar colony formation assay

Lentiviral production

   293T packaging cells were seeded in 100-mm plates at a density of 3.8 ×
   10^6 cells per plate. 293T cells were supplemented with complete DMEM
   (Thermo Fisher Scientific, catalog no. 11966025) containing 10% fetal
   bovine serum (Thermo Fisher Scientific, catalog no. 26140079) and 5%
   penicillin/streptomycin. 293T cells were split three times a week to
   maintain healthy density and incubated at 37°C. Once the optimal number
   of 293T 10-mm plates were produced, they were transfected with psPAX2
   (Addgene, catalog no. 12260), pMD2.G (Addgene, catalog no. 12259), and
   the gene of interest containing pLJM1 lentiviral packaging plasmids
   (Addgene, catalog no. 91980) at concentrations of 1.3, 0.72, and 1.64
   pmol. 293T cells were transfected using the Thermo Fisher Scientific
   Lipofectamine 3000 transfection protocol (Thermo Fisher Scientific,
   catalog no. L3000001). The medium was replaced after 18 hours to permit
   lentiviral production in transfection reagent–free medium.

   Viral medium was harvested after 48, 72, and 96 hours and replaced with
   fresh medium after each harvest. The viral supernatant was filtered
   through 0.45-μm syringe filters Millex-HP Syringe Filter Unit 0.45 μm
   (MilliporeSigma, SLHP033RS) to remove cellular debris and other
   contaminants. NIH-3T3 cells were seeded in 100-mm plates at a density
   of 2.2 × 10^6 cells per plate and supplemented with complete DMEM
   containing 10% calf serum (Thermo Fisher Scientific, catalog
   no.16170086) and 5% penicillin/streptomycin. Once the NIH-3T3 plates
   reached optimal confluency, NIH-3T3 cells were supplemented with the
   viral harvest medium and left to incubate for 24 hours. The viral
   harvest was then removed and supplemented with fresh complete DMEM with
   puromycin (2 μg/ml). The NIH-3T3 cells were incubated for a week under
   these selection conditions until the population density rebounded to
   plate confluency.

Soft agar colony formation assay

   Once the NIH-3T3 cells under each condition were confluent, we
   initiated the soft agar colony formation assay. A 3% agarose solution
   (Guidechem, catalog no. 9012-36-6) was microwaved for 1 min and stored
   in a 45°C water bath to maintain its liquid state. The 3% agarose
   solution was aliquoted and diluted with complete DMEM to yield a 0.6%
   agarose/medium solution. This new mixture was poured into six-well
   plates at 1 ml and allowed to solidify for 30 min. Aliquots of 0.6%
   solution were then diluted with the NIH-3T3 cell medium suspension so
   that the solution was composed of 0.3% agarose and contained a
   concentration of 2 × 10^3 cells/ml. This solution was stored at 37°C
   rather than 45°C to prevent cell death. The solution was plated atop
   the now solid 0.6% agarose layer and allowed to solidify at room
   temperature for 1 hour.

   The agarose suspension of cells was incubated at 37°C for 30 days while
   supplemented with a 1-ml feeder layer of DMEM with puromycin (1 μg/ml)
   and 10% calf serum. After 30 days, the feeder layer was removed, and a
   solution of 0.005% crystal violet dye was plated onto the agarose to
   darken the formed colonies. Colonies were photographed at ×10 and ×20
   magnifications and were counted at 4× magnification.

shTrkB.T1 and neurosphere formation

Real-time polymerase chain reaction

   RNA from Nestin(N)/tv-a;Ink4a/Arf^−/− neural progenitors was isolated
   using the RNeasy Kit (QIAGEN, catalog no. 74104), and 1 μg of total RNA
   was reverse transcribed using the high-capacity cDNA reverse
   transcription kit (Applied Biosystems, catalog no. 4368814). For
   quantitative analysis, we used the PowerUP SYBR Green Master Mix
   (Applied Biosystems, catalog no. A25742) following the manufacturer’s
   instructions. Quantitative polymerase chain reaction (PCR) was
   performed using the QuantStudio 7 Real-Time PCR system. We have used
   previously published primers ([170]60) for TrkB.T1 [AGCAATCGGGAGCATCTCT
   (forward) and TACCCATCCAGTGGGATCTT (reverse)] and TrkB.FL
   [AGCAATCGGGAGCATCTCT (forward) and CTGGCAGAGTCATCGTCGT (reverse)]. The
   threshold cycle number for the genes analyzed was normalized to actin
   [CGTGGGCCGCCCTAGGCACCA (forward) and CTTAGGGTTCAGGGGGGC (reverse)], and
   TrkB.FL and TrkB.T1 levels were normalized to uninfected
   (N)/tv-a;Ink4a/Arf^−/− neural progenitors.

Neurosphere formation

   Neural progenitors isolated from (N)/tv-a;Ink4a/Arf^−/− pups at P0 as
   previously described ([171]33) were infected with RCAS-PDGFB-shSCR
   (scrambled control) or RCAS-PDGFB-shRNA against TrkB.T1, using an
   RCAS-TVA system as has been described previously described ([172]41,
   [173]69). The specific hairpins against TrkB.T1 used in are as follows:
   shTrkB.T1#1
   (GAAAAAGCTAACCACCTGCCCTTTAGATCTCTTGAATCTAAAGGGCAGGTGGTTAGCGGG),
   shTrkB.T1#2
   (GAAAAAGCACTCTCCTCCGCTTTATCTTCTCTTGAAAGATAAAGCGGAGGAGAGTGCGGGGATC), and
   shTrkB.T1#3
   (GAAAAAGTCATAAGATCCCCCTGGATTCTCTTGAAATCCAGGGGGATCTTATGACGGG), along
   with the scrambled control
   (GAAAAAGCTCTACAACCGCTCATCATATCTCTTGAATATGATGAGCGGTTGTAGAGCGGG).

   Briefly, RCAS virus was produced in Chicken fibroblast (DF-1) cells
   maintained with 10% fetal bovine serum (Clonetech, catalog no. 631101)
   in DMEM (Thermo Fisher Scientific, catalog no. 11995-073) with 1%
   penicillin/streptomycin (Thermo Fisher Scientific, catalog no.
   30-002-CI) at 39°C. DF-1 cells were transfected with the indicated RCAS
   plasmid using X-tremeGENE 9 DNA transfection reagent (Roche, catalog
   no. 06365809001) according to the manufacturer’s protocol. RCAS
   transgene expression was confirmed via quantitative PCR analysis. DF-1
   cells were then incubated with murine neurosphere medium for 24 hours
   to package the indicated RCAS plasmid into virus. Our murine
   neurosphere medium consisted of 45 ml of NeuroCult mouse basal medium
   (STEMCELL Technologies, catalog no. 05700), 5 ml of mouse NeuroCult
   proliferation supplement (STEMCELL Technologies, catalog no. 05701), 50
   μl of epidermal growth factor (20 μg/ml; PeproTech Inc., catalog no.
   100-47), 50 μl of basic fibroblast growth factor (10 μg/ml; PeproTech
   Inc., catalog no. 100-18B), 125 μl of heparin (STEMCELL Technologies,
   catalog no. 07980), and 500 μl of penicillin/streptomycin (Thermo
   Fisher Scientific, catalog no. 30-002-CI). This conditioned medium was
   then diluted 1:1 in fresh murine neurosphere medium and used to
   incubate (N)/tv-a;Ink4a/Arf^−/− neural progenitor cells overnight.
   (N)/tv-a;Ink4a/Arf^−/− neural progenitor cells were then washed with
   PBS and maintained in fresh neurosphere medium for 48 hours. They were
   then seeded at a concentration of 50,000 cells per well in ultralow
   attachment six-well plates (Corning, catalog no. 3471). Fresh
   neurosphere medium (0.5 ml per well) was added once a week.
   Neurospheres of >100 μm in diameter were manually counted using a Leica
   DMIL light-emitting diode (LED) microscope with an eyepiece graticule
   at 7, 14, and 21 days, and data were displayed for 21 days.

   Representative bright-field (white-light LED) images of the sphere
   formation assay were obtained using the Dragonfly 200 High-speed
   Confocal Imaging Platform (Andor Technology Ltd.) on a Leica DMi8
   microscope stand with Zyla 4.2 scientific complimentary metal-oxide
   semiconductor camera (Andor Technology Ltd.) and 10×/0.45–numerical
   aperture HC PL APO objective with the software Fusion version 2.3.0.36
   (Oxford Instruments).

Affinity purification–mass spectrometry

   Similar to methods for the colony formation assay, pLJM1 (Addgene)
   constructs containing the NTRK2 inserts of interest (TrkB.FL, TrkB.T1,
   or green fluorescent protein) were transfected into 293T cells, along
   with psPAX and pMD2.G packaging plasmids (Addgene), using
   polyethylenimine (Polysciences). Fresh medium was added 24 hours later,
   and viral supernatant was harvested 24 hours after that. For infection
   of 3T3 cells, 1 × 5^10 cells per well were seeded into six-well plates.
   Lentivirus was used unconcentrated, and 24 hours after seeding, cells
   were infected at a multiplicity of infection of <1. Seventy-two hours
   after seeding, selection was begun for cells successfully expressing
   the constructs using puromycin (2 μg/ml; for 7 days).

   3T3 cells were expanded after selection to create a stable line and
   collected for HA-tag coimmunoprecipitation (Co-IP) analysis (Pierce
   HA-Tag IP/Co-IP Kit, Thermo Fisher Scientific, catalog no. 26180) in
   M-Per buffer (M-PER Mammalian Protein Extraction Reagent, catalog no.
   78501) 3 weeks after infection. This kit includes immobilized antibody
   resin and improved protocols over other methods with a GST-PI3K-SH2-HA
   fusion protein as positive control, is robust and specific, and has
   elution conditions compatible with subsequent downstream analysis. To
   remove detergents needed for Co-IP before downstream MS analysis,
   proteins were resolved by SDS–polyacrylamide gel electrophoresis (PAGE)
   (NuPAGE 10% bis/tris; LifeTech) according to XCell Sure Lock
   Mini-Cell’s guidelines. Gels were stained with SimplyBlue SafeStain,
   and bands were excised and submitted to the Fred Hutch Proteomics Core
   for MS.

   Gel slices were washed with water, 50% acetonitrile/50% water,
   acetonitrile, and ammonium bicarbonate (100 mM), followed by 50%
   acetonitrile/50% ammonium bicarbonate (100 mM). The solution was
   removed, and the gel slices were dried in a speed vac. The gel slices
   were reduced with dithiothreitol (10 mM in 100 mM ammonium bicarbonate)
   at 56°C for 45 min. The solution was removed and discarded. The gel
   slices were alkylated with 2-chloroacetamide (55 mM in 100 mM ammonium
   bicarbonate) and incubated in the dark at ambient temperature for 30
   min. The solution was removed and discarded. The gel slices were washed
   with ammonium bicarbonate (100 mM) for 10 min on a shaker, and an equal
   amount of acetonitrile was added and continued to wash for 10 min on a
   shaker. The solution was removed and discarded, and the gel slices were
   dried in a speed vac for 45 min. The gel slices were cooled on ice, and
   a cold solution of trypsin (12.5 ng/μl; Promega, Madison, WI) in
   ammonium bicarbonate (100 mM) was added, enough to cover the gel slice.
   After 45 min, the trypsin solution was removed and discarded, and an
   equal amount of ammonium bicarbonate (50 mM) was added and incubated
   overnight at 37°C with mixing. Samples were spun down in a microfuge
   and the supernatants were collected. Peptides were extracted from the
   gel slices by adding 0.1% trifluoroacetic acid (TFA) enough to cover
   the slices and mixed at ambient temperature for 30 min. An equal amount
   of acetonitrile was added, and the samples were mixed for an additional
   30 min. The samples were spun on a microfuge and the supernatants were
   pooled. The supernatants were concentrated in a speed vac. All samples
   were desalted using ZipTip C[18] (Millipore, Billerica, MA) and eluted
   with 70% acetonitrile/0.1% TFA. The desalted material was concentrated
   in a speed vac.

   The generated peptide samples were brought up in 2% acetonitrile in
   0.1% formic acid (20 μl) and analyzed (18 μl) by liquid
   chromatography–electrospray ionization tandem MS (MS/MS) with a Thermo
   Fisher Scientific Easy-nLC 1000 (Thermo Fisher Scientific, Waltham,
   MA), coupled to a tribrid Orbitrap Fusion (Thermo Fisher Scientific,
   Waltham, MA) mass spectrometer. In-line desalting was accomplished
   using a reversed-phase trap column (100 μm by 20 mm) packed with Magic
   C[18]AQ (5-μm 200-Å resin; Michrom Bioresources, Auburn, CA), followed
   by peptide separations on a reversed-phase column (75 μm by 250 mm)
   packed with ReproSil-Pur 120 C[18]AQ (3-μm 120-Å resin; Dr. Maisch,
   Germany) directly mounted on the electrospray ion source. A 45-min
   gradient from 2 to 35% acetonitrile in 0.1% formic acid at a flow rate
   of 300 nl/min was used for chromatographic separations. A spray voltage
   of 2200 V was applied to the electrospray tip, and the Orbitrap Fusion
   instrument was operated in the data-dependent mode. MS survey scans
   were in the Orbitrap [ACG (automatic gain control) target value,
   500,000; resolution, 120,000; injection time, 50 ms] with a 3-s cycle
   time, and MS/MS spectra acquisition were detected in the linear ion
   trap (AGC target value, 10,000; injection time, 35 ms) using
   higher-energy C-trap dissociation (HCD) activation with a normalized
   collision energy of 27%. Selected ions were dynamically excluded for 45
   s after a repeat count of 1.

   Data analysis was performed using Proteome Discoverer 2.5 (Thermo
   Fisher Scientific, San Jose, CA) using Sequest HT as the protein
   database search algorithm. The data were searched against a Uniprot
   Mouse (UP000000589 from 7 March 2021) database that included common
   contaminants (cRAPome 29 January 2015). Searches were performed with
   settings for the proteolytic enzyme trypsin, and maximum missed
   cleavages were set to 2. The precursor ion tolerance was set to 10
   parts per million, and the fragment ion tolerance was set to 0.5 Da.
   Variable modifications included oxidation on methionine (+15.995 Da)
   and carbamidomethyl on cysteine (+57.021 Da). All search results were
   run through Percolator for peptide validation, and peptide results were
   filtered to a 1% false discovery rate.

Acknowledgments