Introduction

Although Brain-Derived Neurotrophic Factor (BDNF) is a single molecular message, a mature homodimeric ligand (1), orchestrates many different neurobiological processes, from neurodevelopmental patterning to synaptic plasticity underlying learning and memory (2, 3). While the canonical model suggests that the receptor expression profile dictates the response (4, 5), the broad range of functional outcomes across the nervous system that are attributed to a single ligand implies that the ligand-receptor interaction alone is insufficient to explain the broad functional interpretations of the BDNF message.

The interplay between the receptors TrkB and p75NTR is widely viewed as a regulatory switch that tunes the BDNF signaling response. Upon ligand binding, TrkB autophosphorylates to initiate core Ras/MAPK, PI3K/Akt, and PLC-gamma cascades (6), yet these pathways are qualitatively and quantitatively modified by p75NTR co-expression. This interaction can synergistically enhance TrkB affinity (7–10), prolong signaling duration by reducing receptor ubiquitination (11), or independently activate NF-kappa-B and JNK/c-Jun stress-response pathways (2). However, because these mechanistic ’rules’ were largely defined in transfected cell lines or restricted primary lineages, it remains unclear whether receptor stoichiometry alone can predict signaling outcomes across the heterogeneous cell populations of the nervous system. While single-cell signaling heterogeneity is a recognized feature of other receptor tyrosine kinase (RTK) systems (12), a systematic mapping of how cell identity and maturation state filter the BDNF message is currently unexplored.

This knowledge gap is primarily due to the reliance on biochemical techniques, such as immunoblotting, which measure an aggregate output that is blind to cellular heterogeneity. While these bulk readouts have provided a thorough profile of the BDNF signaling cascade, they cannot uncover how a single ligand might be differentially interpreted across non-uniform subpopulations (13). Newer single-cell technologies offer higher resolution but present contrasting trade-offs: single-cell RNA sequencing effectively maps cell identity but fails to capture the rapid, post-translational phosphorylation events that define the immediate BDNF response (4). Similarly, while standard fluorescence-based flow cytometry provides single-cell signaling data, it is constrained by spectral overlap, preventing the simultaneous measurement of the broad repertoire of phosphoproteins and receptor profiles required to decode a complex signaling landscape (14).

To resolve this decoding problem, we employed highly multiplexed single-cell mass cytometry to construct a high-resolution time course of BDNF signaling across neural lineages and maturation states. By simultaneously measuring receptor composition, receptor surface abundance, and a broad panel of downstream phosphoproteins, our data show a highly filtered signaling landscape where responsiveness is an emergent property of cell identity rather than a simple function of receptor abundance. Our results demonstrate that BDNF signaling is governed by a multi-layered gating system: (1) receptor levels initialize the potential for pathway induction, (2) a sustained reduction of surface TrkB serves as predictor of increased phosphorylation of downstream signaling effectors, and (3) cell-intrinsic environments act as the final arbiter of BDNF sensitivity. By moving beyond the ’average’ signal of bulk biochemistry, we show that even identical receptor profiles yield distinct signaling profiles across cell types, reframing BDNF sensitivity as a form of prepared competence. This multi-layered regulation provides a framework for understanding how cellular context dictates the interpretation of the neurotrophic message.

Results

Single-cell mapping resolves a continuous maturation gradient and lineage diversity

To determine how individual neural identities respond to a uniform BDNF stimulus, we utilized single-cell mass cytometry to simultaneously quantify 19 signaling effectors across primary spinal cord cultures (E14 rat) (Fig. 1A, table S1-2). We selected E14 rat spinal cord as our model system because it provides a developmental window in which neural progenitors, neurons of varying maturity, and non-neuronal populations coexist within a single preparation. Thus, the heterogeneity of this culture system enabled our simultaneous assessment of BDNF signaling across multiple lineages and maturation states.

Fig. 1.

Leiden clustering of identity marker expression distinguished 20 unique cell populations (Fig. 1B, fig. S1A), which were partitioned into four primary categories based on marker expression level (fig. S1A, B): Progenitor (Sox2+, 12.8%), Neuronal (ßIII-tubulin+, 52.5%), Non-Neuronal (17.0%), and Unassigned (17.6%). Unassigned cells represent those that could not be unambiguously defined by our panel of identity markers.

Notably, based on expression levels of neuronal identity markers, we were able to resolve a developmental trajectory originating from Sox2+ progenitors (Cluster 1), through intermediate immature neurons (Clusters 7, 2, 13), to the most mature NeuN-hi neurons (Cluster 4) (Fig. 1C, Fig. S1D). We identified additional Non-Neuronal cell types, including nNOS+ fated progenitors, astrocytes (s100ß+), oligodendrocytes (OligoO4+) and microglia (CX3CR1+) based on their distinguishing identity marker (Fig. 1C, Fig. S1A, B). Microglia can be further classified based on the presence (“laden”) or absence (“unladen”) of neuronal identity markers, which may represent the phagocytic state of these CX3CR1+ cells. By resolving this maturation gradient and the cell-type heterogeneity at single-cell resolution, we established a high-dimensional system to test to what extent BDNF signaling is a universal event or a state-dependent property of specific lineages.

The global BDNF response is driven by a discrete subpopulations of cells

To determine if our platform recapitulated well-established canonical signaling profiles, we first aggregated our single-cell data into a “pseudobulk” format to assess the global response to BDNF (fig. S2-3). As expected, BDNF induced rapid activation across core Ras/MAPK and PI3K/Akt pathways, including pERK, pAkt, and pCREB, relative to controls where BDNF deprivation was continued (Fig. 2A, fig. S4A). These responses were largely abrogated by the Trk inhibitor K252a (Fig. 2B, fig. S4A). In addition, BDNF stimulation exhibited a unique phosphoproteomic response which was distinct from the general trophic rescue elicited by refeeding with rich medium (“Rescue”) (fig. S5), confirming the specificity of the BDNF stimulus.

Fig. 2.

However, single-cell analysis indicates that these aggregate trends obscure highly heterogeneous responses. Even at peak activation (15 min), the population did not respond uniformly, but instead discrete subpopulations, ranging from 47% to 75% of cells, responded (Fig. 2C, fig. S4B, table S3). This observation provides evidence that BDNF responsiveness is not a uniform population event but is restricted to specific cellular contexts (Fig. 2D).

Maturation state and cell identity regulate BDNF signaling competency

We next determined if the heterogeneity of BDNF responsiveness was structured by cell identity, for instance if neuronal cells responded and non-neuronal cells did not. While both neuronal and non-neuronal cells show increased activation of established pathways (fig. S6A), neuronal cells exhibit a greater magnitude of change (fig. S6B). Inhibition of Trk activity with K252a significantly reduces BDNF-induced activation to control or near-control levels in both cell type categories (fig. S6C).

A more granular assessment of cell-type-specific pathway activation revealed a diversity of responsiveness (Fig. 3A-B, fig. S7A). Intriguingly, we observed a maturation-dependent acquisition of signaling competence along the neuronal lineage. Sox2+ progenitors remained relatively signaling-silent (1.7-fold pERK) compared to intermediate and mature neurons, which exhibited a 16.2- to 17.2-fold induction at the same time point (Fig. 3C, table S4).

Fig. 3.

Crucially, progenitors showed robust activation when treated with rescue media (Fig. 3D, table S4), confirming that they are intrinsically signaling-competent but specifically insensitive to BDNF. Further comparison of treatment-dependent responses shows a relatively consistent response to rescue media along the neuronal lineage, suggesting that maturation-dependent responsiveness is not generalizable to any stimulation (fig. S8).

Non-neuronal lineages similarly showed divergent sensitivity: astrocytes and “laden” microglia were highly responsive, whereas “unladen” microglia remained signaling-silent (Fig. 3E). Furthermore, while the Trk-inhibitor K252a blunted most activated pathways, the failure to completely abolish pERK induction in several lineages (Fig. 3F, fig. S7B, table S4) suggests that Trk activity is not the sole determinant of the signaling patterns, hinting at additional cell-intrinsic regulatory layers.

Unsupervised clustering reveals cell identity-dependent BDNF signaling signatures

By measuring 19 signaling markers at once, we were able to define signaling patterns in identified cell types based on the temporal changes of individual marker abundance. However, assessing each marker separately limits our analysis and does not take advantage of our multi-dimensional dataset. Thus, before exploring the observation of Trk-activity dependence, we expanded on our characterization of responses to BDNF by defining “signaling signatures” based on input from all measured signaling markers without regard to cell type.

To move beyond separate analysis of individual markers, we performed unsupervised Leiden clustering based exclusively on signaling marker abundance, agnostic to cell identity (Fig. 4A). This re-clustering analysis resolved 21 signaling clusters (fig. S9A) distinguished by their marker expression profiles (fig. S9B). By plotting cells back onto the generated signaling UMAP by time and treatment, we were able to observe shifts in the distribution of cells to define signaling signatures. Based on the shift in distribution upon BDNF stimulation (Fig. 4B), we identified three temporal signaling signatures – Signature A (rapid pAkt/pERK), Signature B (sustained pS6), and Signature C (complex multi-marker activation) – as well as Basal and Non-Signaling Signatures (Fig. 4C). Of the 21 signaling clusters identified by the clustering analysis, 13 were represented in these defined signaling signatures, each exhibiting a unique signaling marker expression profile (Fig. 4D).

Fig. 4.

Surprisingly, when we mapped cell types back onto the signaling UMAP, they grouped together, even without identity markers in the input data (Fig. 4E). When we plotted the distribution of individual cell types over time, each followed a unique temporal profile. Intermediate neurons progressed primarily through Signature B, whereas mature neurons adopted the more complex Signature C (Fig. 4F). Although these temporal snapshots represent population shifts rather than individual cell trajectories, the divergence between mature and intermediate neurons shows that maturation state defines not just the magnitude, but the “signature” of BDNF signal induction. These data suggest that BDNF responsiveness is a programmed feature of cell identity, where different lineages decode the same ligand into unique, temporally regulated signaling signatures (Fig. 4G).

The sustained reduction of surface TrkB correlates with BDNF sensitivity

We next sought to identify the molecular mechanisms governing these distinct signaling signatures. A primary candidate for this variability is the relative abundance of the two BDNF receptors themselves; however, it remains unknown to what degree TrkB and p75NTR density dictate signaling magnitude and pathway specificity across cell populations.

To resolve this, we first quantified the contribution of total TrkB receptor expression using a tiered pseudobulk analysis, where each cell was categorized based on its total level of TrkB (Fig. 5A; “None”, “Lo”, “Med”, “Hi”). While 91.23% of cells express some level of TrkB, receptor density was not a consistent predictor of signaling output (Fig. 5B, fig. S10A). Increased activation of effectors downstream of BDNF-TrkB, such as pERK and pS6, were primarily associated with TrkB presence; however, pCREB was not directly correlated with increased TrkB levels, suggesting TrkB abundance is not the only factor driving its activation (Fig. 5B). Cells not expressing TrkB (“None”) showed no activation of certain signaling markers, such as pNF-kappa-B and p-c-Jun, though this population of cells did exhibit increase in pERK, pS6 and pCREB (Fig. 5B), supporting the established idea that BDNF signaling is not solely mediated through TrkB. Additionally, the increase in pERK in the “None” cell population aligns with our observation that Trk inhibition only partially suppresses ERK activation (Fig. 2C).

Fig. 5.

To assess whether different levels of TrkB influenced signaling profiles in different cell types, we incorporated TrkB expression level categories into our signaling signature analysis (Fig. 5C). In cell types with robust responses, such as intermediate and mature neurons, all levels of TrkB in the given cell type were present in BDNF-responsive (insets; Signatures A-C) and Basal Signatures (blue dashed outline). Within signaling signatures, there is some grouping based on TrkB level – most notably, TrkB-Hi in mature neurons in Signature C (green dashed outline) – suggesting some contribution of TrkB level to the temporal signaling profile. Although fewer in number, some progenitor cells were present in BDNF-responsive signatures (inset), which had varying levels of TrkB. However, the majority of progenitors exhibit a Basal Signature (blue dashed outline in Fig. 5C), despite harboring Med-to-Hi levels of TrkB (Fig. 5D). This presents an apparent paradox whereby TrkB-expressing cells (i.e. progenitors) can be insensitive to BDNF.

One possible mechanism for the TrkB-high, signaling-silent progenitors is that TrkB in progenitors might not undergo endocytosis, a process thought to be required for BDNF signaling. Using a surface-specific staining protocol, we quantified ligand-induced TrkB depletion from the plasma membrane (Fig. 5E). Our results show a clear correlation between signaling competence and a reduction of surface receptor abundance (Fig. 5F). While intermediate and mature neurons showed significant surface TrkB depletion at later time points, unresponsive progenitors exhibited stable surface profiles. This lack of surface TrkB reduction provides a correlative explanation for why receptor-positive cells remain functionally signaling-silent. These data indicate that the reduction of surface receptors over time is an important part of BDNF regulation (Fig. 5G), suggesting that surface receptor dynamics may serve as a predictor of responsiveness beyond static protein abundance alone.

To assess the contribution of p75NTR expression on BDNF-induced signaling, we performed a tiered pseudobulk analysis based on the level of p75NTR in each cell. Comparison of expression level categories shows an apparent shift in signaling patterns: high p75NTR density was associated with increased stress-responsive p-c-Jun and p38 activation, but also unexpectedly amplified trophic markers like pCREB and pERK (fig. S10B). Taken together, these analyses suggest that receptor presence alone, regardless of density or receptor profile, is insufficient to confer signaling competence, which is predicted by the sustained time-dependent reduction of surface TrkB.

Cell identity modulates signaling response set by receptor stoichiometry

Our results show that the abundance of each BDNF receptor individually contributes to the BDNF response but does not account for all observed signaling changes. This is expected, as others have established that interplay between TrkB and p75NTR can tune signaling responses (7–11). We therefore wanted to understand how receptor stoichiometry influences the profile of BDNF-induced signaling.

To address this question, we first compared signaling responses in cells with different combinations of TrkB and BDNF. By plotting the level of p75NTR vs total TrkB expression, we were segmented four categories of cells based on our previously-applied tiered pseudobulk analysis: p75NTR–/TrkB–, p75NTR+/TrkB–, p75NTR–/TrkB+ and p75NTR+/TrkB+ (Fig. 6A). Consistent with synergistic models, we observed that TrkB and p75NTR act in concert to amplify markers like pERK, pS6, pCREB, and pNF-kappa-B, while p75NTR expression alone – in the absence of TrkB – was sufficient to drive p-c-Jun activity (Fig. 6B, fig. S11).

Fig. 6.

We next asked how receptor expression levels influence BDNF-induced signaling responses within a given cell type. Using astrocytes as an example, we observe synergistic amplification of certain signaling markers, where p75NTR+/TrkB+ astrocytes exhibited high activation of pERK and pS6 with a milder increase in pAkt (fig. S12). Other signaling marker activity did not show synergy: pGSK3b exhibited similar responses in TrkB-med cells regardless of p75NTR presence, and p-p53 was only increased in p75NTR–/ TrkB-med, suggesting a TrkB-dependent/p75NTR-independent response. Together, this analysis shows that receptor stoichiometry can influence BDNF-induced signaling in the same cell type.

To determine if receptor expression level is the ultimate determinant of signaling output, we the analyzed signaling responses across different cell types, holding the receptor expression profile constant. We isolated subpopulations of astrocytes, progenitors, intermediate neurons, and mature neurons that were matched for identical surface levels of both TrkB and p75NTR (Fig. 6C). Surprisingly, an identical receptor profile (for instance TrkB-med/p75(+)) did not result in an identical response. Instead, each cell type produced distinct signaling patterns (Fig. 6D). This observation held when comparing cell-type specific responses in a different receptor profile (TrkB-lo/p75(+); Fig. 6E, fig. S13). Additional comparison of the same cell type across these two receptor profiles highlights our observation that receptor stoichiometry can modulate signaling responses.

However, these results ultimately demonstrate that while receptors are necessary for BDNF responsiveness, they are not sufficient to dictate the signaling pattern (Fig. 6F). Instead, the receptor stoichiometry sets the potential of BDNF sensitivity, which is further tuned by lineage-dependent regulators. These observations suggest that BDNF signaling is a hierarchical process whereby the cell’s internal state determines the signaling outcome of this single extracellular cue.

Discussion

The signaling response to BDNF is traditionally understood through receptor biology. Our results support a more holistic model in which receptor stoichiometry sets the potential for BDNF sensitivity, and the intrinsic environment allows for a cell to engage that potential. By moving beyond the “average” signals of bulk biochemistry, our high-dimensional analysis reveals a BDNF-induced signaling landscape where identical receptor profiles produce differing signaling patterns across various cell types. By reframing BDNF sensitivity as a form of prepared competence, we can view BDNF-induced signaling as interconnected points of regulation (Fig. 7). While receptor stoichiometry primes a cell to respond to BDNF, additional factors – such as surface receptor dynamics – that are intrinsic to individual cell types determine exactly when and how the BDNF message is translated into a functional outcome.

Fig. 7.

Cell type-specific context defines BDNF response set by receptor stoichiometry

A prevailing assumption of receptor-ligand signaling posits that receptor ratio and density completely predict signaling outcomes. However, by isolating subpopulations with matched receptor expression patterns across unique lineages, we demonstrate that this model is incomplete. To our surprise, identical receptor profiles yielded largely different signaling programs across cell types: a mature neuron’s response is fundamentally different from that of an astrocyte or progenitor even at the same TrkB/p75 stoichiometries. This indicates that the interpretation of a BDNF signal is not invariantly set by the receptor stoichiometry itself but is modulated by the intracellular context. Our results expand upon computational models of NGF/TrkA signaling which suggest that protein expression noise – specifically intrinsic noise (15, 16) – drives single-cell ERK variability (12, 17). Furthermore, our multiplexed measurements provide a concrete example of a biological context that sets the ’cell state’ described in these models: the extent of neuronal differentiation sets a maturation gradient with distinct BDNF responsiveness. This suggests that previously observed stochastic variance may actually reflect programmed, deterministic differences across neuronal maturity, thus reframing ’noise’ as a developmentally regulated feature.

Sustained reduction of surface TrkB is a functional correlate of responsiveness

If receptor density cannot predict signaling output, does receptor trafficking serve as the primary indicator of responsiveness? Following ligand binding, TrkB is internalized into signaling endosomes to trigger canonical trophic cascades (6, 18). Our data indicate a correlation between signaling competence and receptor trafficking: while neurons and astrocytes show significant ligand-induced surface TrkB depletion, non-responsive populations – most notably progenitors – exhibit stable surface receptor levels. Similar to other signaling systems modulated by receptor internalization (19), our observation suggests a trafficking-based explanation for why receptor-positive cells remain signaling-silent. While these findings are correlative and do not yet prove a causal requirement for internalization, the consistency of this pattern across neural lineages suggests that trafficking kinetics are a more reliable predictor of BDNF sensitivity than static protein abundance. Defining the precise molecular regulators of this internalization at single-cell resolution remains a key open question to be addressed by future endocytic perturbation studies.

Furthermore, the progenitor “paradox” (i.e. high TrkB expression paired with negligible internalization) may represent a point of signaling regulation rather than an inability to respond. In this context, progenitor non-responsiveness could be a mechanism of “prepared competence,” where cells maintain a high TrkB density as they await a developmental window that permits differentiation. A compelling candidate for this gate is a developmental switch from truncated (TrkB-T1) to full-length (TrkB-FL) isoform expression. Under this hypothesis, TrkB-T1 would act as a signaling sink or dominant-negative regulator (20, 21) until TrkB-FL expression becomes dominant, which would have profound consequences for progenitor cell-fate decisions (22, 23). Although our antibody targets the shared extracellular domain and cannot discern these isoforms, our data are consistent with this hypothesized cell-intrinsic factor due to the presence of the receptor but a lack of sustained surface reduction and signaling response. Thus, further investigation is warranted into this potential mechanism of prepared competence.

Receptors are necessary, but not sufficient in BDNF signaling regulation

While receptor trafficking provides a rationalization of BDNF responsiveness, how does the interplay between TrkB and p75NTR shape BDNF signaling? Our results suggest that BDNF pathways utilize a combinatorial logic where the level of receptor availability tunes specific cascades. Consistent with established synergistic models (7, 9, 21, 22), we observe that TrkB and p75NTR act synergistically to amplify trophic pathways like pERK and pNF-kappa-B, while p75NTR alone is sufficient to drive stress-responsive p-c-Jun activity. We posit that these patterns may be further regulated by p75NTR cleavage dynamics (23), adding another layer of receptor-mediated regulation. This high-resolution mapping highlights that p75NTR is not merely a “death receptor” but a critical synergistic partner that heightens cellular sensitivity to the BDNF message in the context of our experimental system.

It is unsurprising that receptors are necessary to initiate BDNF-induced signaling; however, our results indicate that receptor presence alone is not sufficient to dictate the final output. We demonstrate that this receptor interplay is modulated by cellular context unique to each lineage. This is evidenced by our finding that equivalent receptor expression patterns yield varying signaling programs (Fig. 6), that only a discrete subpopulation (∼47–75%) of total cells respond to BDNF stimulation (Fig. 2), and that unsupervised clustering based solely on signaling markers independently groups cells by their biological identity (Fig. 2).

Intrinsic factors ultimately gate BDNF interpretation across distinct cell types

While we cannot determine from fixed data whether individual cells traverse signaling signatures sequentially or preferentially occupy different signatures, insights from these data – made possible through high-dimensional single-cell proteomics – suggest a hierarchy of filters. These intrinsic filters may include receptor isoform ratio (20, 21), cleavage and internalization dynamics (23), cell cycle phase (24), or the epigenetic landscape (25–27), which likely determines the accessibility of target genes for BDNF-induced expression. Such multifactorial logic – where the signaling pattern rather than amplitude encodes biological meaning – mirrors findings in other systems, including calcium oscillations (28–31), TNF-NF-kappa-B dynamics (32–34), Notch signaling (35), growth factor-induced PI3K activity (36), and Wnt signaling (37, 38). By demonstrating that lineage modulates stoichiometry, we reframe BDNF signaling as a tiered process where maturation state defines the functional meaning of the extracellular cue.

While our high-dimensional approach provides a new framework for understanding BDNF signaling, several limitations of the current study warrant consideration. First, our findings regarding TrkB internalization and downstream signaling remain primarily correlative. Although the lack of receptor trafficking in progenitors serves as a consistent predictor of signaling silence, future investigation is required to establish a direct causal requirement for internalization in this specific context. Second, our analysis relies on discrete temporal snapshots. While these timepoints capture the evolution of signaling signatures across the population, they cannot directly track the dynamic trajectories of individual cells as they transition between signaling states. Furthermore, our findings are derived from an in vitro primary culture system; while this allowed for the high-resolution mapping of endogenous spinal cord lineages, future studies are required to determine how these nodes of regulation are modulated within the complex three-dimensional environment of the in vivo niche. Finally, our use of an antibody targeting the shared extracellular domain of TrkB introduces ambiguity regarding receptor isoforms. While we hypothesize that the “progenitor paradox” may be driven by a developmental switch between truncated (TrkB-T1) and full-length (TrkB-FL) isoforms, definitive confirmation of this “active gate” will require the development of isoform-specific proteomic probes. Addressing these limitations through targeted genetic perturbations will be essential to further validating the multi-layered model of neurotrophin regulation suggested here. Additionally, our study exclusively examines the response to mature BDNF. Because proBDNF preferentially engages p75NTR rather than TrkB (39–41), comparing the signaling landscapes elicited by each ligand form would directly test whether the regulatory nodes we describe are reconfigured when the receptor entry point shifts from TrkB to p75NTR, potentially identifying an additional layer of ligand-form-dependent route diversity.

Beyond fundamental signaling, these findings suggest a new logic for neurotrophic drug development for neurological diseases and disorders. Our model predicts that global TrkB agonists will exhibit cell-type-specific efficacy in vivo, potentially explaining the inconsistent results and partial success of BDNF-based therapies in clinical trials (42, 43). Effective strategies must likely move beyond simple receptor agonism toward combination approaches that shift the intracellular context – transitioning a cell from a signaling-silent state to a BDNF-sensitive one. For example, the use of maturation-stage-specific HDAC inhibitors could remodel the epigenetic landscape of stagnant populations (25, 44), lowering the “intrinsic filter” to permit TrkB-mediated differentiation. However, the transition in vivo faces the formidable challenge of cell-type-specific delivery (43), as ensuring context-shifting agents reach the intended subpopulation without inducing off-target effects in already-responsive neurons remains a hurdle. Ultimately, this study suggests that neurotrophin sensitivity is a form of prepared competence that lies not just in the ligand and the receptors, but in the internal machinery of the cell itself.

Materials and methods

1. Animals

All experiments were conducted in accordance with AAALAC guidelines on a protocol approved by the University of Virginia ACUC. Timed pregnant Sprague Dawley rats were purchased from Charles River (Strain Code: 001) at embryonic day 12 (E12) and housed on a 12hr light/dark cycle prior to embryonic tissue dissection.

2. Cell culture

Primary dissociated neural cultures were prepared from E14 Sprague Dawley rat spinal cords. Briefly, spinal cords (sacral level excluded) were dissected in chilled HBSS, minced, and enzymatically digested for 15 min at 37°C in a filtered enzyme solution containing 0.25% (w/v) trypsin, 0.04% (w/v) hyaluronidase, and 0.04% (w/v) DNase. Tissue was mechanically dissociated by trituration in plating media (Basal Medium Eagle supplemented with 5% FBS, 1X N2 Supplement, 1X GlutaMAX, and 1% Penicillin-Streptomycin) and passed through a 70 µm cell strainer. Cells were plated at a density of 5×105 cells/well in 12-well plates pre-coated with 0.1 mg/mL poly-D-lysine (overnight) and 15 µg/mL laminin (>2hr) in PBS. Initial cell viability was >90% as determined by Trypan blue exclusion prior to plating. Within 16hrs, plating media was replaced with Maintenance Media (Neurobasal media supplemented with 1X B27 and 1X GlutaMAX). Cultures were maintained at 37°C in 5% CO2, with half-volume media exchanges every 48hrs.

At day in vitro (DIV) 3.5, cultures were washed three times with a working volume (0.5mL) of Deprivation Media (Neurobasal and 1% Penicillin-Streptomycin) and deprived of trophic support for 12hrs prior to treatment. Cells were then stimulated with 50 ng/mL BDNF (9), 200 nM K252a (45, 46), or a combination in deprivation media for specified durations (5min to 72hrs). Maintenance Media was used as a Rescue positive control, while fresh Deprivation Media was applied to the Continued Deprivation negative control.

3. Mass cytometry sample collection

Following stimulation, treatment media was replaced with 500 µL of room-temperature Accutase. Cells were monitored for detachment, mechanically dissociated via P1000 trituration into a single-cell suspension, and filtered through a 70 µm strainer into 500 µL of 4% paraformaldehyde (PFA) in PBS. Total processing time from media removal to fixation was maintained under 90s to preserve transient signaling states as best as possible. After a 10 min fixation at room temperature, cells were pelleted (500*g, 5 min), resuspended in 500 µL of Cell Storage Media (CSM; DPBS with 0.5% w/v BSA and 0.2% w/v sodium azide), and aliquoted for quality control and cell counting. Samples were stored at -80°C until analysis.

Each biological replicate (n) represents a neural culture generated from an independent timed-pregnant litter. To ensure statistical robustness and account for batch effects, treatments were collected with 2-4 biological replicates per time point (table S2), with BDNF-treated samples (5min to 16hr) performed in triplicate and baseline (0hr) controls in quadruplicate.

4. Sample barcoding, staining, and mass cytometry

Samples were stained according to established high-dimensional mass cytometry protocols (47–51). To minimize inter-sample staining variability and mitigate batch effects, individual samples were uniquely barcoded using a “6-choose-3” combinatorial system of palladium metals (52, 53). To account for potential inter-run variations across barcoded batches, a “universal reference” sample – comprising pooled aliquots (11µL, ∼2.44%) of all experimental samples pooled – was included in every barcode set. Barcoding was achieved through reversible permeabilization in 0.02% saponin for 15 min at room temperature with agitation (54).

Following barcoding, samples were pooled and blocked with 10% Normal Donkey Serum in CSM for 30min. C were first stained for 30min with a primary antibody cocktail targeting extracellular epitopes, including the TrkB extracellular domain that enabled our assessment of surface TrkB dynamics (table S1). Samples were then fixed and permeabilized with ice-cold 100% methanol for 10min, with vortexing every 2min to prevent cell clumping. Permeabilized cells were subsequently stained for 1hr with a cocktail of antibodies against intracellular targets (table S1). To distinguish single-cell events from debris, samples were incubated overnight at 4°C in 1.6% PFA/DPBS containing 125 µM Iridium (Ir)-intercalator (Standard Biotools, South San Francisco, CA).

Prior to analysis, pooled samples were resuspended in Maxpar Cell Acquisition Solution (CAS) Plus and filtered through a 40µm strainer. Data were acquired on a Helios CyTOF XT system (Standard Biotools) at an acquisition rate below 500 events per second. Samples were processed in four sequential batches of 20 barcoded samples each (table S2). Data were collected using Helios CyTOF.

5. Normalization, debarcoding, and single cell gating

Raw FCS files were standardized via bead-based normalization to account for instrument sensitivity fluctuations (55). Samples were then debarcoded using a “6-choose-3” combinatorial system to assign events to their respective biological replicates (53). To ensure high data quality for the resulting dataset, single-cell events were isolated through a standardized cleanup gating strategy in Cytobank.

Events were first filtered by low barcode separation distance and high Mahalanobis distance to remove poorly assigned cells. Singlets were isolated by comparing event center to event length and width. Residual EQ calibration beads were excluded by gating out high cerium (Ce140) events, while environmental contaminants (Sn120, I127, CS133, Ba138, Pb208) were identified and removed by gating against the Iridium DNA intercalator (Ir191/193). A specific gate on Gd157 was applied to eliminate bleed-through signal from Nd141. The resulting high-fidelity 680,000-cell dataset of single-cell events were exported as individual FCS files per sample for downstream analysis (https://community.cytobank.org/cytobank/experiments/121988).

6. Batch correction

To account for potential batch effects, data from the “universal reference” sample included in each barcode batch were used to normalize signal intensities across batches using a quantile-based adjustment method (https://github.com/CUHIMSR/CytofBatchAdjust; R-4.5.2) (56). This approach effectively minimized technical variance (batch effects) while preserving biological heterogeneity across the longitudinal dataset. Although 72hr samples were collected and processed through the batch correction, they were not included in downstream analyses in the present study due to the possibility of post-treatment secretion of growth factors confounding our interpretation of results.

7. Leiden clustering and UMAP visualization

Single-cell events were partitioned into molecularly defined subpopulations using the Leiden community detection algorithm (Python 3.12.4) (49, 57, 58). We employed a two-stage clustering strategy to decouple biological identity from signaling state. First, cells were clustered using lineage-specific ID markers (table s1) to define cell-type identities (Fig. 1B, fig. S1). Second, to identify unsupervised signaling signatures (Fig. 4), events – excluding those classified as ‘Unassigned’ – were re-clustered based solely on phosphoprotein markers without a priori identity information. Lineage assignments were subsequently mapped back to signaling clusters using unique event identifiers to determine the cell-type composition of each signature.

High-dimensional data were dimensionally reduced and visualized via Uniform Manifold Approximation and Projection (UMAP) (59, 60) using the following parameters: nearest neighbors = 15, metric = Euclidean, local connectivity = 1, epochs = 1000. All UMAP visualizations were generated in R-4.5.2.

8. Data analysis and quantification

Pre-processed mass cytometry data were analyzed using custom R (R-4.5.2) scripts. To quantify signaling induction across the heterogeneous population, we defined a “signaling response” as any event exceeding the 95th percentile of the baseline signal intensity observed in the 0-hour (trophic-deprived) control for each respective phosphoprotein. This thresholding was applied globally and, where indicated, to individual cell-type clusters to account for lineage-specific baseline variances (fig. s2).

Responsiveness was quantified as percent abundance, representing the proportion of “responsive” cells relative to the total population within a given treatment condition or cell-type cluster. For stoichiometry analysis, receptor expression tiers (“None”, “Lo”, “Med”, “Hi”) were established via logical biaxial gating of total TrkB and p75NTR across the entire dataset. This tiered approach allowed for the identification of the “progenitor paradox” and supported the finding that receptor density alone is an unreliable predictor of signaling output. All custom analysis scripts and data processing pipelines are available at https://github.com/TooSewellForSkool/Sewell-et-al-BDNF-Signaling-Manuscript.

9. Statistics

Data are presented as mean±SD. Each biological replicate (n) represents a neural culture derived from an independent timed-pregnant litter. For comparisons between two groups, an unpaired two-tailed Student’s t-test was performed compared to 0hr. All statistical analyses were conducted using R-4.5.2, and an alpha level of p < 0.05 was considered statistically significant.

Data availability

Raw FCS files are available at https://community.cytobank.org/cytobank/experiments/121988. All custom analysis scripts and data processing pipelines are available at https://github.com/TooSewellForSkool/Sewell-et-al-BDNF-Signaling-Manuscript.

Additional files

Supplementary material. Fig. S1. Characterization of cell identity clusters and their distribution. Fig. S2. Defining the cells that are responsive to treatment. Fig. S3. Comparison of different threshold cutoffs and different treatments. Fig. S4. BDNF-induced responses and the effect of K252A inhibition – Extension of Fig. 2B, C. Fig. S5. Pseudobulk signaling response to rescue and BDNF treatment. Fig. S6. BDNF-induced signaling changes in Neuronal and Non-Neuronal cell categories. Fig. S7. BDNF-induced signaling across distinct cell types – Extension of Fig. 3B, F. Fig. S8. Comparison of signaling responses across treatments in neuronal lineage cell types. Fig. S9. Signaling signatures agnostic to cell ID clustering information. Fig. S10. BDNF receptor level contribution to pseudobulk signaling response. Fig. S11. Pseudobulk analysis of receptor level contribution to signaling heterogeneity. Fig. S12. Receptor expression level contributes to signaling heterogeneity within the same cell type – Extension of Fig. 6B. Fig. S13. Signaling comparison of different cell types with the same receptor profile – Extension of Fig. 6D, E. Table S1. Mass cytometry antibody panel. Table S2. Sample organization into barcode sets. Table S3. p-values for statistics performed in Fig. 2C and Fig. S4B. Table S4. p-values for statistics performed in Fig. 3

Additional information

Funding

HHS | National Institutes of Health (NIH) (F31NS135970)

• Jonathon M Sewell

HHS | National Institutes of Health (NIH) (T32GM139787)

• Jonathon M Sewell

HHS | National Institutes of Health (NIH) (R01NS111220)

• Eli R Zunder

HHS | National Institutes of Health (NIH) (R01NS083378)

• Bettina Winckler

HHS | National Institutes of Health (NIH) (R01NS136342)

• Christopher D Deppmann

UV | Brain Institute, University of Virginia (UVA Brain Institute) (Presidential Fellowship in Collaborative Neuroscience)

• Jonathon M Sewell