Multiple-scale neuroendocrine signals connect brain and pituitary hormone rhythms

Significance

The hypothalamo–pituitary axis controls a wide range of homeostatic processes, including growth, stress, and reproduction. Despite this fact, the hypothalamic neuron firing patterns that lead to slowly evolving pituitary hormone rhythms remain enigmatic. Here, we used in vivo amperometric recordings in freely behaving mice to investigate how tuberoinfundibular neurons release dopamine (DA) at the median eminence (ME) to control pituitary prolactin secretion. Using this approach, we show that DA release occurs as multiple locally generated and time-scaled secretory events, which are integrated over a range of minutes across the ME. These results provide a broad physiological mechanism for the dialogue that occurs between the brain and pituitary to dictate hormone rhythms over multiple timescales, from ultradian to seasonal.

Abstract

Small assemblies of hypothalamic “parvocellular” neurons release their neuroendocrine signals at the median eminence (ME) to control long-lasting pituitary hormone rhythms essential for homeostasis. How such rapid hypothalamic neurotransmission leads to slowly evolving hormonal signals remains unknown. Here, we show that the temporal organization of dopamine (DA) release events in freely behaving animals relies on a set of characteristic features that are adapted to the dynamic dopaminergic control of pituitary prolactin secretion, a key reproductive hormone. First, locally generated DA release signals are organized over more than four orders of magnitude (0.001 Hz–10 Hz). Second, these DA events are finely tuned within and between frequency domains as building blocks that recur over days to weeks. Third, an integration time window is detected across the ME and consists of high-frequency DA discharges that are coordinated within the minutes range. Thus, a hierarchical combination of time-scaled neuroendocrine signals displays local–global integration to connect brain–pituitary rhythms and pace hormone secretion.

A remarkable function of the brain is its capability to integrate temporal information with complex physiological responses. This has been well established for behavioral responses such as nonrapid eye movement (NREM) sleep, where three neuronal oscillations with distinct frequency bands support information transfer (1). However, the neuronal mechanisms that orchestrate the dialogue between the brain and other basic functions like reproduction, lactation, and growth remain largely unknown (2–5). They depend on the fine tuning of pituitary hormone pulses by small assemblies of hypothalamic neuroendocrine or parvocellular neurons that release specific secretagogues at the median eminence (ME) (4, 6).

Here, we took advantage of the anatomical organization of the ME to investigate how the tuberoinfundibular (TIDA) neuronal population (7, 8) releases dopamine (DA) to negatively regulate pituitary secretion of prolactin (PRL), a key reproductive hormone (2). To do so, miniaturized amperometric carbon fiber implants were used to detect DA release events (9) for days to weeks in freely behaving mice. Using this approach, we uncovered a hierarchically organized delivery of release events over four orders of magnitude (from <0.1 s to several hours), which correlate with the dynamics of PRL in the bloodstream.

Results

Frequency Coding of DA Release Events in Vivo.

To characterize the release dynamics of TIDA nerve terminals in vivo, we used long-term constant voltage amperometry in awake mice using thin (30-µm tip diameter) carbon fibers implanted into the ME (Fig. 1A). Voltage was clamped at −700 mV to allow detection of DA released from TIDA neurons. DA amperometry was performed continuously during several days, and the relationship with PRL secretion was assessed using tail blood microsampling for high-sensitivity mouse (m)PRL ELISA developed in-house (10) (Fig. 1A). Single carbon fiber electrode recordings revealed robust DA currents [median 325 nA, interquartile range (IQR): 127–822 nA] due to oxidation of DA to dopamine-o-quinone (Fig. 1A), and these could be robustly detected over the long term (Fig. 1B) (n = 7 virgin female mice). We then used DA as a relevant readout to explore the dynamics of TIDA neuron population function in freely behaving animals. DA currents at the ME level discharged over different timescales (Fig. 1 B and C) and more frequently during the night than day (Fig. 1D) (mean counts per hour from zeitgeber (ZT) 0, in 6 h blocks: 18.7, 27.2, 28.6, and 30.4), implying that the strength of TIDA neuron excitability is likely modular around the day/night switch. DA release events were often grouped and interspaced by long-lasting (dozens of minutes to several hours) silent periods, suggesting nested relations between high- and low-frequency output patterns (Fig. 1 B and C). No clear association between DA current density and estrus cycle stage was detected (Fig. 1C and Fig. S1).

Fig. 1.

In vivo monitoring of DA release events at the median eminence (ME) level. (A) Electrodes were implanted at the ME of mice and dopamine (DA) was detected using constant voltage amperometry. Serial blood microsampling was performed from the tail vein. (B) Representative 24-h recording of DA release (Top, shaded area is lights out), with zoom of a 10-min sequence (Bottom). (C) Representative 11-d recording from a female mouse. Each vertical line corresponds to a single secretion event. The stage of the estrus cycle is indicated on the Left for each day (M, metestrus; D, diestrus; P, proestrus, E, estrus). (D) Mean distribution of DA release events during the day (n = 80 d from seven female mice). (E) Histogram of interevent intervals (IEIs): two prominent frequencies are apparent at 1.5 and 12 Hz (n = 80 d, from seven female mice). (F) Relation between DA and PRL. Average normalized PRL levels occurring around a DA event (n = 501 DA events, from six 1-h long sessions) (black, mean; blue, SEM). (G) DA secretory response to an i.p. injection of 1 µg ovine PRL (PRL injected at time 0) (from five animals, seven injections). (H) Distribution of the IEIs of DA events induced by i.p. injection of PRL. (I) Example of simultaneous recording of PRL levels (red) and DA release events (black). In all cases, bar graphs show the mean ± SEM.

Fig. S1.

Average frequency of DA release events during the different stages of the estrus cycle (M, metestrus; D, diestrus; P, proestrus, E, estrus) (n = 5 female mice, shaded area is lights out). No statistical difference between the different days of the cycle was detected (P > 0.05, mixed effects model). Individual data points, which are outside of the interquartile range (indicated by bars), are shown as dots.

Analysis of interevent intervals (IEI) for DA release unveiled a wide range of time intervals, from less than 100 ms to a few hours, with two principal frequencies of 1.5 Hz and 12 Hz (Fig. 1E). Elevated release from TIDA neurons corresponded with periods of lowered PRL concentration (Fig. 1F). A delay of several minutes between decreasing PRL levels and the onset of high-frequency DA release events was also observed following exogenous PRL injection (Fig. 1G), and these occurred at similar frequencies (0.9 Hz and 17 Hz) (Fig. 1H) to those recorded during spontaneous activity (Fig. 1E). Notably, this response outlasted the decrease in PRL levels (Fig. S2), supporting a role for persistent PRL receptor signaling (2, 11, 12) or other mediators (13–15) in the generation of high-frequency DA release events.

Fig. S2.

Levels of PRL detected after i.p. injection of 1 µg ovine (o)PRL in B6 female mice (n = 6, mean ± SEM).

Conversely, the arrest of high-frequency DA release events was followed a few minutes later by an increase in PRL levels (Fig. 1 F and I), resembling the previously described responses to administration of a D2 receptor antagonist (10). Thus, the TIDA neuron population has the capability to generate bouts of DA release events at relatively high frequencies (hertz range), which are inversely correlated with PRL levels in the bloodstream of freely moving mice. We were also able to record these episodic high-frequency DA events over a number of days during lactation (Fig. S3), although their amplitude and frequency were lower, most likely due to the reported loss of DA granular content at this time (9).

Fig. S3.

(A) Density of secretory events in a mouse recorded during lactation (days 6–16) and weaning (days 1–4) (n = 2). Insets show examples of the raw signal. (B–D) Distribution of IEIs during different phases of the recording in A.

Long-Range Organization of DA Release Events at the Local ME Level.

We next examined whether these subsecond DA release events possess a secondary/tertiary organization at the local level, i.e., in the close vicinity of carbon fiber tips. Using cluster analysis to group DA currents on the basis of their shape, and bootstrapping to identify temporal series of events appearing with a higher-than-chance frequency during the recording period, a specific distribution could be revealed. In six of the seven recorded mice, several repetitive patterns of DA release events were found, with stereotypical features remaining consistent between different animals recorded on different days (Fig. 2A and Fig. S4).

Fig. 2.

Temporal patterning of DA release events. (A) Distribution of IEIs for each class of event obtained after clustering all events from one recording by shape (n = 13,541). Insets show average event shape in each group; the proportion of each class is shown near each graph. (B–E) Example of temporal patterns of DA release. Each dot represents a single DA release event, colored depending on the subgroup (as in Fig. 2A). Each line shows one repetition of the sequence during the recording; five examples of repetition are shown for each pattern. (F–I) Frequency of the four temporal patterns during 8 d of recording.

Fig. S4.

Properties of release patterns. (A) Choice of statistically significant patterns. Each recording was analyzed for n-event long (3 ≤ n ≤ 10) patterns, repeated at least five times. The number of occurrences of each pattern are plotted against n, each dot representing one temporal pattern. The same procedure was repeated on 1,000 computer-generated sequences, with the same number of events, same distribution of IEI, and same proportion of different classes. A 95% confidence limit (blue dotted line) was then calculated for the distribution of the maximum number of repeats in the bootstrap samples. Patterns repeated a greater number of times than the bootstrap limit are indicated in green and are considered to be occurring with a higher-than-chance probability. (B) Distribution of temporal patterns during the day (n = 7 mice). (C) Duration of statistically significant temporal patterns in the recording from mouse 4, in relation to the number of events in the pattern. (D) Duration of statistically significant six event-long temporal patterns in seven different mice.

Further analyses demonstrated that these stereotyped patterns of DA release were not randomly distributed, but rather appeared as chains of sequential events within the same group and/or between groups (Fig. 2 B–E). These recurrent motifs of DA release events were scaled from the millisecond (Fig. 2 B–D) right up to the hour (Fig. 2E) range, and could even be detected over days (Fig. 2 F–I). Thus, the mechanisms controlling TIDA neuron activities appear to be inherently robust.

Local–Global Integration of DA Release Events Across the Median Eminence.

A long-standing question regarding parvocellular neuron function is how nerve terminals discharge their neurohormones across the ME to sculpt pituitary output (2–4, 6). Given that TIDA nerve terminals abut over the whole ME (7, 8), dual-carbon fiber recordings were carried out 500 µm apart rostrocaudally, spanning the population (n = 3 animals). Whereas distant DA events at high frequencies (≥1 Hz) were not synchronized (Fig. 3A), DA events were coordinated with IEIs in the minutes range during most of the recordings (Fig. 3 B and C). This temporal coordination was not seen when each electrode was considered separately (Fig. 3D), suggesting that it is not simply due to hypothalamic PRL feedback, but rather effects on TIDA neuron interactions. Frequencies of 1.39 ± 0.12 Hz and 10.08 ± 2.6 Hz were both present during the dual electrode recordings of coordinated DA release events (n = 6) (Fig. 3 E and F). Such spatial organization strengthens the view of a large-scale coordination within the TIDA neuron population, which may provide a means for transforming short-lived hypothalamic signals into long-lasting inputs for downstream endocrine targets.

Fig. 3.

Spatial patterning of DA release events. (A) Representative double recording of DA secretion at distant sites in the ME (500 µm rostrocaudal), showing desynchronization of release events at the minute timescale. (B) Distribution of events from a double recording. (Top) “Rug plot” of DA release events, where each vertical line represents one event detected by one of the two electrodes. (Bottom) Density plot of the DA events, showing coincidence over a long timescale. (C) Cross-distribution of IEI between the two electrodes, showing reciprocal delays during a ∼7-min time lag. (D) Autocorrelation of the signals on each of the single electrodes shows only the expected peak at lag 0, suggesting that the coordination is not dependent on pituitary feedback. (E and F) Distribution of IEIs from the signals detected by the two electrodes during a period in which DA release was only detected by one electrode (E) (light purple) or during a period of coordination between both electrodes (F) (light purple and green; dark purple shows overlap of IEIs between both electrodes).

Discussion

Our results show how an ensemble of parvocellular TIDA neurons delivers its neuroendocrine products toward ME portal vessels in a freely behaving mouse model. DA release events are repeated over weeks as network-driven rhythms that cover more than four orders of magnitude of frequency, from infraslow (<0.001 Hz) to fast rhythms (1–10 Hz). This organization occurs not only locally within, but also across the TIDA neuron assembly, as DA release events are scaled over the minute range throughout the ME (Fig. 4).

Fig. 4.

Schematic of the brain–pituitary dialogue proposed to underlie hypothalamic dopaminergic control of pituitary prolactin secretion. Illustrated are three subsets of hypothalamic TIDA neurons (colored in green, brown, and magenta), which each locally release DA at the median eminence level (where the first loop of portal capillaries reside). Local DA release events are organized in the frequency domain (0.001 Hz–10 Hz) and recur as sequences. Local–global integration across the median eminence coordinates high frequency DA release events within the minutes range. This allows the build-up of DA in the portal blood required to efficiently inhibit pituitary prolactin secretion.

Specifically, the use of miniaturized carbon fibers stereotaxically implanted into the ME allowed us to detect and discriminate DA-related currents in vivo, which were far more complex, but also more organized than spike firing activities recorded in parvocellular neurons from either brain slices (9, 14–17) or anesthetized animals (18). Even though the small tip of the carbon fiber was likely able to detect DA release from only a few TIDA neurons, we observed a variety of rhythms. First, high-frequency (about 1 and 10 Hz) events were prominent locally but not synchronized globally. As the site of recording is variable and these rhythms were observed in all animals, a large number of local DA release processes presumably originate from TIDA neurons capable of secreting at high rates. The latter would be considered as “executive” in the top-down control of pituitary PRL rhythms by hypothalamic DA inputs, because they coincided with drops in pituitary PRL secretion. Second, slower rhythms of DA release (with time periods of minutes to hours) were detectable locally due to the ability of small carbon fibers to measure DA events over days to weeks with no noticeable deleterious effects. Strikingly, these were not distinguishable from high-frequency DA events with which a hierarchal combination occurred regarding both the specific frequencies generated and how they organize in time as temporal motifs. Because the local–global integration of high-frequency DA events occurred over frequencies of one or more minutes across the ME, slow rhythms may orchestrate the delivery of longer, but highly ordered DA outputs from the TIDA neuron assembly toward the pituitary responder.

The current study performed in freely behaving animals poses the question of how the TIDA neuronal network generates such a hierarchal organization of DA release events in vivo. Whereas classical PRL feedback (2) is able to account for a proportion of the high-frequency DA release events through direct stimulation of TIDA neuron electrical activity, it cannot explain slower rhythms, including those organized over a minutes range across the ME. This observation raises the possibility that both intranetwork modes of information transfer (14, 19) and neuronal inputs (13–16), which were recently revealed in acute brain slice studies, may contribute to the coding of DA release at the ME level. Nonetheless, the present study suggests that the TIDA neuron network has the inherent capability to code DA release over time periods consistent with the pacing of PRL secretion.

Finally, it has recently been shown that local somatodentritic DA release from the TIDA population is able to influence close neuronal neighbors within the arcuate nucleus (19). As D1 and D2 receptors are expressed in the ME (20), DA release events at this location, even those organized over slow rhythms, may also contribute to the regulation of other neurohormones, such as those underlying circhoral luteinizing- and growth-hormone pulses (20, 21).

The discovery of a multiple-timescale integration of DA delivery at the neurohemal space provides a hitherto unknown element concerning how the brain dialogues with peripheral organs via a neuroendocrine connection. Such hierarchical organization of rhythms has been observed in other brain regions where multiple oscillations co-occur, with the slower oscillation generally driving local, faster oscillations (1). A similar multiple-timescale neurohemal code may plausibly be shared by other assemblies of hypothalamic parvocellular neurons. Notably, the ME is capable of delivering hormone changes over a wide range of timescales, from ultradian to seasonal (22, 23). This neurohemal structure may thus provide a model system for investigating how parvocellular outputs are translated into slowly evolving endocrine outcomes such as reproduction, growth, metabolism, and stress.

Materials and Methods

Detailed methods are provided in SI Materials and Methods. All animal procedures were approved by the local ethical committee under agreement CEEA-LR-12185 according to European Union Directive 2010/63/EU. Briefly, carbon fiber microelectrodes were fabricated using a single 30-µm thread of carbon fiber, coated in Nafion and connected to a gold-plated pin. C57/BL6 female mice were stereotaxically implanted with carbon fiber microelectrodes at the level of the median eminence [stereotaxic coordinates (relative to Bregma) −1.3 mm rostrocaudal, 0 mm mediolateral, and 6.1 mm ventral]. After recovery, mice were transferred to recording cages, connected to an electrical swivel to allow for free movement, and carbon fibers were held at 700 mV throughout the recording to detect secretion of DA. Repeated tail blood microsampling was performed to measure blood PRL levels, using a home-made ELISA. All statistical analysis was performed with R software.

SI Materials and Methods

Animals.

C57/BL6 virgin female mice were housed under a 12 h light/dark cycle (lights on at 0900), with ad libitum access to water and food. The estrus cycle of the animals was checked daily between 9 and 10 AM through examination of vaginal cytology. Pregnancy and lactation were induced using standard timed mating procedures. All animal procedures were approved by the local ethical committee under agreement CEEA-LR-12185 according to European Union Directive 2010/63/EU. Because this study included only one experimental group of animals, no randomization or blinding was required.

Fabrication of Carbon Fiber Microelectrodes and Optic Fibers.

A single thread of carbon fiber (30-μm diameter; World Precision Instruments) was inserted under a stereomicroscope through the opening of a flexible silica capillary (Polymicro Technologies; 40 µm i.d., 150 µm o.d., Molex), leaving ∼500 μm protruding from each extremity. A gold-plated pin (World Precision Instruments) was then fixed to one end using conductive glue (carbon-epoxy, World Precision Instruments). The carbon epoxy was allowed to cure for 2 h at 75 °C, then cyanoacrylate glue was used to seal the capillary at the detecting end, taking care not to cover the tip of the carbon fiber. A Nafion coating was applied by dipping the exposed fiber in Nafion perfluorinated solution (Sigma-Aldrich), while applying a 3-V potential to the microelectrode. This step strongly reduces the nonspecific signal derived from the oxidation of ascorbate molecules (24). Before implantation, electrodes were calibrated in vitro using standard solutions of DA, and those that did not respond to DA (e.g., because of faulty contact between the pin and the fiber, broken fiber, etc.) were discarded.

Implantation of Microelectrodes.

Mice were anesthetized with 10 µL/gram body weight of a mix of 1% ketamine and 0.1% xylazine in 0.9% NaCl. The head of the mouse was then attached to a stereotaxic frame and sterilized with 10% (wt/vol) betadine. A sagittal incision was made through the skin of the cranium to expose the sagittal suture before craniotomy, using a dental burr. Three jeweler’s screws (Plastics One) were fixed to the skull to improve stability of the head-cap, and a ground pin was attached to one of them. A support guide cannula (Plastics One) was then inserted 1.5 mm above the median eminence and fixed to the skull with Dentalon dental acrylic (Phymep). A carbon fiber microelectrode was then slowly passed through the guide cannula, so that its tip reached the median eminence at the stereotaxic coordinates (relative to Bregma) −1.3 mm rostrocaudal, 0 mm mediolateral, and 6.1 mm ventral. The implant was finally blocked with dental acrylic and the mouse was left to recover for at least a week before initiating the recording. Postoperative analgesia was provided using i.m. injection of the nonsteroidal antiinflammatory ketoprofen. A similar procedure was used for double implants, except that fibers were fixed to a double cannula, with 500-μm spacing between the two guides. The gold-plated pins were isolated using silicone tubing to prevent electrical contact between the two channels, and a drop of dental wax was used to isolate the two pins.

In Vivo Amperometry.

Mice were transferred to recording cages, connected to an electrical swivel (Plastics One) to allow for free movement. A Faraday cage was used to limit electrical noise. Food and water were provided ad libitum. Recordings were started at least 2 days following transfer to the recording cage to allow the mouse to habituate to the new environment. Throughout, carbon fiber microelectrodes were held at 700 mV using a HEKA EPC10 amplifier, as previously reported (2) to detect secretion of DA, and oxidation currents were recorded at 1 kHz. Animals for which no electrical signal was detected or those where the electrode was found to be implanted away from the ME at the end of the experiment were discarded from the analysis.

Prolactin Measurements.

Repeated tail blood microsampling was performed during 1-h sessions in the late afternoon, when the density of DA signal was higher. Briefly, 4-µL blood samples were withdrawn from the tail vein every 5 min and diluted with 96 µL PBS-Tween 20 before storage at −80 °C pending analysis. Prolactin concentrations were measured using a home-made ELISA, as previously reported (10). In some experiments, i.p. injection of ovine PRL (Sigma) was performed during tail-tip blood sampling (Fig. S2).

Data Analysis.

Raw data from PatchMaster (HEKA) files were imported into IGOR Pro (Wavemetrics) and a threshold algorithm was used to identify peaks. Peaks were defined as having intensity greater than 10 SDs from baseline noise, with a minimum width of at least 50 ms. The times and shapes of the detected peaks were then exported to flat text files and further analyzed using R software. Event distribution was studied using standard techniques for the analysis of point processes, such as analysis of the interspike intervals. Classification of DA release events was performed using k-means clustering. The width, amplitude, area under the curve (AUC), and five quantiles of the peak shape were used as classifiers for clustering.

Repeated temporal patterns were found by first determining all n-long sequences of events (3 ≤ n ≤ 10), with a maximum total length of 5 h, taking into account the cluster to which each group belonged. Patterns formed by the same series of events (same sequence of clusters, and with the same IEIs, with a tolerance of ±15% for each IEI), were grouped together. Groups with fewer than five sequences were not considered. The same process was repeated on 1,000 bootstrap samples of the time series, obtained by randomly shuffling the IEI. The 95th percentile of the distribution of the maximum number of repeated patterns found in a group of each bootstrap sample was used as the limit at which results were deemed statistically significant. As expected, this limit is higher for shorter sequences (Fig. S4A).

The relation between PRL levels and DA events was calculated by averaging the linearly interpolated PRL levels measured around each detected DA event. The levels were then normalized between 0 and 1 and averaged between recordings. This generated the “mean” PRL response to a DA release event shown in Fig. 1F, demonstrating the inverse relationship between DA and PRL.

For double recordings, cross-density histograms were generated by plotting the histogram of the IEI between the events recorded by the two electrodes, as previously reported (9, 25). Furthermore, a similarity index [cosine similarity, defined as (A*B)/(||A||*||B||), where A and B are the vectors of events at the two electrodes] was used to determine whether the “loose synchronization” of the events in the two recordings was higher than that expected by chance. The similarity index was calculated for each recording, and for 1,000 bootstrap samples in which the IEIs from a single electrode were randomly shuffled. In all cases, the real data had similarity above the 95% confidence interval of the bootstrapped data.

Estrus cycle data were analyzed using a mixed effects model. The cycle stage and the time of the day (in 4-h blocks) were used as fixed effects, the animal was used as a random effect. Tukey’s all-pair comparisons post hoc analysis was then performed to compare the effect of the stage of the cycle.