Inhibiting mPFC OXTR neurons in female mice decreases male choice in a task of sociosexual decision-making.

(A) Experimental groups virally expressing imaging and optogenetic constructs in the right mPFC. Female Cre-expressing mice expressed GCaMP6m in pyramidal neurons and either Halorhodopsin (Halo) or a control fluorophore (control) in OXTR neurons (n = 8, 7 respectively). One additional fluorophore-injected subject was Cre-negative and was therefore included in the control group (n = 8 total). A GRIN lens was implanted directly above the expression site for optical imaging and optogenetic light delivery. PL: prelimbic cortex. (B) Behavioral task, in which freely-moving subjects chose between social interaction or a sweetened milk solution on each trial. Within a given session, subjects were exposed to both a novel adult male and adolescent female (each for half the total number of trials, typically 16 out of 32). (C) Trial structure, including timing of optogenetic light delivery. (D) Calculation of male and female behavioral preference in each session. Male preference is the proportion of trials that the subject chose the male stimulus while he was present in the arena (typically 16 trials). Female preference is the proportion of trials that the subject chose the adolescent female stimulus while she was present in the arena (typically 16 trials). (E) Female versus male preference over control (left, n = 40) and Halo (right, n = 42) sessions before the introduction of optogenetic light (pre-opto. baseline sessions). A binomial logistic regression model was fit to each dataset (overlaid curves; fixed effect of female preference (mean-centered); control: Estimate = 4.007, SE = 0.610, Z = 6.57, P < 0.001***; Halo: Estimate = 2.192, SE = 0.646, Z = 3.40, P < 0.001***). (F) Female versus male preference over control (left, n = 36) and Halo (right, n = 39) sessions with optogenetic light delivery (opto. sessions). Overlaid are baseline fits from (E), which were used to test whether male preference during opto. differed from baseline (actual male preference during opto. versus predicted male preference from baseline fit). (G) Actual - predicted male preference in control and Halo opto. sessions (same n as (F)). Session values transformed to logit scale to match binomial generalized linear mixed model (GLMM) used for statistical testing. Boxplots show median and interquartile range (IQR) over sessions. Only Halo group showed a significant difference between actual and predicted male preference (control: Estimate = -0.097, SE = 0.110, Z = -0.88, P = 0.380; Halo: Estimate = -0.225, SE = 0.107, Z = -2.11, P = 0.035*). In control group only, actual and predicted male preference were significantly equivalent within a tolerance range of +/- 1 male trials (Two One-Sided Tests (TOST) procedure: P control = 0.025*, P Halo = 0.213). (H) Actual - predicted male preference on non-receptive (NR, n = 23) and sexually receptive (SR, n = 16) Halo opto. sessions. Boxplots show median and IQR over sessions. Estrous state (SR vs. NR) modulated the difference between actual and predicted male preference (binomial GLMM with estrous state as fixed effect; Estimate = -0.329, SE = 0.190, Z = -1.73, P (one-sided) = 0.042*). (I) On each trial, position was projected onto goal axis and aligned to start of directed movement towards male or milk goal. (J-K) Projected position (J) and corresponding velocity (K) relative to alignment point on male and milk trials in control (left) and Halo (right) opto. sessions. Thick lines and shaded areas show median and IQR over pooled trials (control: n = 173 male, 402 milk; Halo: n = 162, 461). (L) AUC of velocity on male versus milk trials from (K). Vertical dotted line indicates run start (0.6 s before alignment point).

Inhibiting mPFC OXTR neurons increases pyramidal population activity.

(A) Cell maps in example control and Halo sessions (n = 99 and 131 cells, respectively; median of 101 cells per session over all sessions). (B) Activity traces of 10 example cells highlighted in (A). Optogenetic light overlaid in orange. (C) Proportions of cells in each session that significantly increased (black) or decreased (grey) their activity in response to optogenetic light turning on (Wilcoxon signed-rank test over male and milk trials in each session, P < 0.05 after Benjamini-Hochberg correction; n = 36 control, 39 Halo sessions here and (D-F, L)). All remaining cells (not shown) showed no significant change. Right: Proportions of modulated cells (increased or decreased activity) in control versus Halo sessions (same n as Left). Halo group showed higher proportion of modulated cells (binomial GLMM with group as fixed effect: Estimate = 3.101, SE = 0.418, Z = 7.43, P < 0.001***). (D) Population response to optogenetic light turning on (male and milk trials) in control and Halo sessions. Halo group showed stronger boost of population activity compared to control (LME with group as fixed effect: Estimate = 0.745, SE = 0.183, t(13.95) = 4.07, P = 0.001**). (E) Population activity immediately before (left) and at the start (right) of optogenetic light in control and Halo sessions (LME with group (control, Halo), and timepoint (pre-light, light start) as fixed effects; significant group x timepoint interaction (F(1, 132.11) = 51.87, P < 0.001***). Groups only differed at the start of optogenetic light (pre-light (control - Halo): Estimate = -0.040, SE = 0.112, t(25.8) = -0.35, P = 0.727; light start (control - Halo): Estimate = -0.884, SE = 0.112, t(25.8) = -7.87, P < 0.001***; Holm-corrected P values). (F) Population response to optogenetic light turning off (male and milk trials) in control and Halo sessions. Halo group showed stronger drop of population activity compared to control (LME with group as fixed effect: Estimate = -0.815, SE = 0.238, t(13.99) = -3.43, P = 0.004**). (G) Representative reconstructed pair of pyramidal (PN; top) and Halorhodopsin-expressing OXTR (OxtrN; bottom) neurons in L2/3 mPFC. Both neurons were fluorescently labeled and reconstructed using biocytin-streptavidin staining. (H) Membrane potential recordings of neuron pair in (G) during three 6-s long experimental phases. (I) Number of action potentials in each phase for 3 pairs (Top: PN, Bottom: OxtrN). Pair from (G-H) shown in black. (J) Trial timepoints, which were either individual time samples (vertical lines) or windows containing multiple time samples (bubbles). Within windows, activity was averaged over samples to obtain one value per window. Orange bar indicates optogenetic light, which could extend through 1.8 s (from run start) on at least some trials (Figure S6B). (K-L) Neural activity over trial timepoints in two example sessions (K) and all sessions (L). Activity plotted separately on male (left) and milk (right) trials. (K) Activity of individual cells (rows) in example sessions, sorted by difference in activity between male and milk trials at goal timepoint. (L) Comparison of population activity between all control and Halo sessions over time. Compared to control group, population activity in Halo group was significantly elevated at timepoints with optogenetic light. The only exception was the “Tone, Light start” timepoint on male trials, where the P value was trending for significance after correcting for multiple comparisons (LME with group (control, Halo), timepoint, and trial type (male, milk) as fixed effects; significant group x timepoint x trial type interaction (F(15, 2190) = 2.51, P = 0.001**), followed by post hoc group comparisons at each timepoint; black bar indicates P < 0.05 after Holm correction for multiple comparisons). Throughout the figure, boxplots and thick lines/shaded areas show median and IQR over sessions.

Inhibiting mPFC OXTR neurons delays the ability of a specific ensemble to represent the male option.

(A) Definition of separability score. In each session, the ability of each cell to discriminate between male and milk trials was evaluated at the goal timepoint using ROC AUC analysis. Each cell’s AUC was rescaled between -1 to 1 to define a separability score. A positive value indicated preference for the male, a negative value indicated preference for the milk, and a value of 0 indicated no preference. (B) Distribution of separability values over cells in control (left, n = 36) and Halo (right, n = 39) sessions (same n throughout figure). (C) Approach to define male-representing (MALE) cells in each session, involving pairwise correlating the activity of cells that were sorted and binned by their separability value (2.5 percentile-width bins; 40 bins total). Activity was taken on male trials. This approach produced a matrix of Spearman correlation coefficients for each session. (D) Summary coefficient matrix on male trials in control group (median over sessions). MALE cells were detected from this matrix using change point analysis (80th percentile, Figure S7A). OTHER cells were defined as all remaining cells. (E) Comparison of within-group correlation coefficients between MALE and OTHER cells in control sessions (each line is one session). Coefficients were Fisher-transformed for statistical testing. MALE cells showed more strongly correlated activity compared to OTHER cells (LME with MALE - OTHER coefficients as outcome variable: Estimate = 0.117, SE = 0.033, t(6.89) = 3.52, P = 0.00997**). (F-G) Same as (D-E) but now in Halo sessions. MALE and OTHER cells did not differ in their strength of correlated activity (LME with MALE - OTHER coefficients as outcome variable: Estimate = 0.038, SE = 0.053, t(7.00) = 0.72, P = 0.497). (H-I) Activity of MALE and OTHER cells on male (left) and milk (right) trials in control (H) and Halo (I) sessions. (J) Activity of MALE (left) and OTHER (right) cells on male and milk trials in control sessions. MALE cells significantly and stably distinguished between male and milk trials starting from 0.6 s before the run start, whereas OTHER cells did not distinguish trial outcome at any timepoint (LME with male - milk activity as outcome variable, and timepoint and cell group (MALE, OTHER) as fixed effects; significant timepoint x cell group interaction (F(15, 1085) = 49.14, P < 0.001***), followed by post hoc tests (difference from 0) at each timepoint in each cell group; black bar indicates P < 0.05 after Holm correction for multiple comparisons). (K) Same as (J), but now in Halo sessions. MALE cells briefly distinguished between male and milk trials at 0.6 s before the run start but only became stably significant starting from 0.4 s into the run. OTHER cells did not distinguish trial outcome at any timepoint (LME with male - milk activity as outcome variable, and timepoint and cell group (MALE, OTHER) as fixed effects; significant timepoint x cell group interaction (F(15, 1178) = 15.19, P < 0.001***), followed by post hoc tests (difference from 0) at each timepoint in each cell group; star and black bar indicates P < 0.05 after Holm correction for multiple comparisons). Throughout the figure, boxplots and thick lines/shaded areas indicate median and IQR over sessions.

Computational model reproduces effects of inhibiting mPFC OXTR neurons on disrupting male choice and MALE cells’ representation before the run.

(A) Model architecture, consisting of three cell groups: OXTR neurons, MALE cells, and all other (non-MALE) pyramidal neurons (OTHER cells). OXTR neurons inhibit MALE and OTHER cells, whereas MALE and OTHER cells excite OXTR neurons. MALE and OTHER cells also excite themselves through within-group connectivity, with MALE cells having stronger connectivity (thicker line). To reproduce experimental control and Halo conditions, the model was simulated with normal or reduced input from OXTR neurons (OXTR ON or OFF, respectively). (B) Representative simulation run of MALE and OTHER cells’ activity (n = 500 OXTR ON trials, 500 OXTR OFF trials). Outcome of each trial (male (red) or milk (blue)) determined by probability contours. (C) Experimentally recorded activity of MALE and OTHER cells on male and milk trials, plotted separately for control (n = 36) and Halo (n = 39) sessions. In each session, the activity of each cell group was summarized over male and milk trials separately, giving two values per session. The distribution of activity values over sessions for each trial type was plotted as a dot and boundary box (median and IQR over sessions). (D) Male preference (left) and AUC of MALE cells (right) on simulated OXTR ON and OXTR OFF trials. Horizontal and vertical lines show median and full range over n = 12 simulation runs. (E) Symmetric model architecture containing equally-sized MALE and OTHER cell groups. (F-G) Same as (B, D) but now using symmetric model.

Validation of viral expression and implant placements in mPFC.

(A) Example viral expression and implant location in the right mPFC of subject F995576 (implant locations of all subjects shown in (F)). GCaMP6m expression (green) and Nissl stain (red). PL: prelimbic cortex. Cg1: cingulate cortex, area 1. (B) Example GCaMP6m (green) and mCherry (red) expression in control subject F1113632. DAPI stain (blue). (C) Example GCaMP6m (green) and Halorhodopsin-mCherry (red) expression in Halo subject F1216810. DAPI stain (blue). (D-E) GCaMP6m expression occurs in Neurogranin-positive cells (neural marker) but not glutamate decarboxylase (GAD)-positive cells (GABAergic marker), consistent with pyramidal expression. (D) GCaMP6m (green), Neurogranin (red) and DAPI (blue). Right top: Zoom-in on Neurogranin stain; Right bottom: Zoom-in on GCaMP6m expression. Arrows point to three example Neurogranin-positive neurons expressing GCaMP6m. (E) GCaMP6m (green), GAD (red) and DAPI (blue). Right top: Zoom-in on GAD stain; Right bottom: Zoom-in on GCaMP6m expression. Arrows point to three example GAD-positive neurons that do not express GCaMP6m. (F) Locations of the implant base for all subjects (n = 8 control, 8 Halo). Letters in parentheses (e.g., “(A)”) indicate figure panels showing representative images from these subjects.

Validation of Halorhodopsin-expressing mPFC OXTR neurons.

Current clamp recordings of Halorhodopsin-expressing mPFC OXTR neurons during wash-in experiments with aCSF only (n = 6 cells) or aCSF + oxytocin (OXT; n = 12 cells; 6 shown here and remaining 6 shown in Figure S3) for at least 12 minutes. Each panel shows one neuron and includes: Top left: Membrane potential response to negative and positive current injections (50 pA steps, 3 s). Top right: Validation of inhibitory response to 1.5 s optogenetic light exposure. Thick line and shaded area show mean +/- standard error of the mean (SEM) over 5 repetitions. Bottom left: Histological reconstruction of recorded neuron using biocytin-streptavidin staining. All reconstructed neurons display a stellate-like morphology, consistent with previous work identifying OXTR neurons in mPFC22. No reconstruction available for cells 5, 6, and 10. Bottom right: Raw (black) and low-pass filtered (red) recordings during wash-in experiments. Wash-in began at time 0 and lasted until the last timepoint. Blue shaded areas indicate a “baseline window” (180 s before wash-in starts) and an “active window” (last 180 s of wash-in), used for quantifications.

Validation of Halorhodopsin-expressing mPFC OXTR neurons (Continued).

(A) Continued from Figure S2: remaining 6 Halorhodopsin-expressing mPFC OXTR neurons receiving aCSF + OXT wash-in. (B) Change in membrane potential from baseline to active windows for all neurons exposed to aCSF only (n = 6) or aCSF + OXT (n = 12). For each neuron, the membrane potential was low-pass filtered and averaged over all timepoints within each window to produce one baseline value and one active value. Then the change from baseline to active was calculated. OXT exposure significantly increased the membrane potential compared to aCSF only (Wilcoxon rank-sum test, P < 0.001***). Boxplots show median and IQR over neurons.

Experimental design.

(A) Experiment timeline for each subject. All sessions were performed daily, with the exception of arena habituation, which occurred at the end of the second day of drink training. (B) Trial structure, including definitions of optogenetic light duration and return latency. (C) Distribution of social stimulus presentation order (male or female first) and location (left- or right-hand side of maze relative to start arm) over optogenetics sessions used in behavioral and neural analyses (n = 36 control, 39 Halo sessions). Sessions sorted by presentation order and social side. (D) Representative crystal violet-stained vaginal samples taken at the end of each session. Based on estrous phase, sessions were classified as putatively sexually receptive or non-receptive. (E) Definition of latency to return to start arm at the end of each trial, used as a measure of task performance (lower values indicate better task performance). (F) Summarized return latencies in individual control (n = 8) and Halo (n = 8) subjects over maze testing sessions (median over all trials in each session). Optogenetic light was introduced on session 6 or 7, once a subject’s return latency had been assessed to be stable (see verification in (G)). (G) Same data as (F) but now aligning sessions to the start of optogenetic light (5 sessions before and 5 sessions with optogenetic light; n = 7-8 control and 7-8 Halo subjects at each session index). Thick lines and shaded areas show median and IQR over subjects in each group. During pre-opto. sessions, there was a significant effect of session on return latency, but no significant effect of group (control, Halo) or session x group interaction (LME with session (mean-centered) and group as fixed effects: F_session(1, 14) = 10.02, P_session = 0.007**; F_group(1, 14) = 0.03, P_group = 0.862; F_interaction(1, 14) = 0.39, P_interaction = 0.544). This indicated that return latency improved over pre-opto. sessions, with a similar trajectory in control and Halo groups. During opto. sessions, there were no significant effects (F_session(1, 58.74) = 0.66, P_session = 0.419; F_group(1, 13.97) = 0.80, P_group = 0.386; F_interaction(1, 58.74) = 0.14, P_interaction = 0.707), indicating that the return latency was stable and similar between groups. (H) Duration of optogenetic light in control versus Halo sessions (same n as (C); light duration summarized over all trials within each session to obtain one value per session). Halo group did not differ from control (LME with group as fixed effect: Estimate = -0.127, SE = 0.309, t(13.99) = -0.41, P = 0.686). (I) Weight loss in control versus Halo opto. sessions (same n as (C)). Weight loss computed relative to a baseline measurement before water scheduling (see (A)). Halo group did not differ from control (LME with group as fixed effect: Estimate = -1.413, SE = 0.871, t(13.99) = -1.62, P = 0.127). Boxplots in (H-I) show median and IQR over sessions.

Binomial logistic regression adequately fits pre-opto. sessions, and female preference does not differ between control and Halo groups.

(A) Comparison of female preference between control (n = 40) and Halo (n = 42) sessions before optogenetic light stimulation (pre-opto. sessions; binomial GLMM with group as fixed effect: Estimate = 0.119, SE = 0.246, Z = 0.49, P = 0.628). (B) Actual - predicted male preference in control and Halo pre-opto. sessions (same n as (A); predictions for both groups from pre-opto. regression model fits in Figure 1E). Session values transformed to logit scale to match binomial GLMM used for statistical testing. No significant difference between actual and predicted male preference in either group (control: Estimate = 0.000, SE = 0.086, Z = 0, P = 1; Halo: Estimate = 0.000, SE = 0.083, Z = 0, P = 1). In both groups, actual and predicted male preference were significantly equivalent within a tolerance range of +/- 1 male trials (TOST: P control and P Halo both < 0.001***). (C) Comparison of female preference between control (n = 36) and Halo (n = 39) opto. sessions (binomial GLMM with group as fixed effect: Estimate = -0.027, SE = 0.246, Z = -0.11, P = 0.914). Throughout the figure, boxplots show median and IQR over sessions.

Characterization of trial timepoints in relation to each other and optogenetic light.

(A) Overlap between timepoints on male and milk trials (columns) in control and Halo groups (rows). Within a given plot, each row and column indicate one timepoint. For each row, the proportion of trials where a given timepoint overlaps with itself and each of the other timepoints is indicated by a color from dark blue (0) to yellow (1). In control group, n = 173 male and 402 milk trials (pooled over opto. sessions). In Halo group, n = 162 male and 461 milk trials. Deviation from dark blue outside of the diagonal shows the extent of overlap between timepoints. This deviation was low across trial types and groups, indicating that timepoints were largely and consistently distinct. (B) At each timepoint, proportion of trials of a given type (male or milk) containing optogenetic light. Proportions were computed on trials from each session (n = 36 control sessions and 39 Halo sessions). Dots show individual sessions. Thick lines and shaded areas show median and IQR over sessions in each group.

Detection and characterization of MALE cells.

(A) To detect MALE cells, the upper-right triangle of the correlation matrix in Figure 3D was taken. Each row was summarized (median, excluding diagonal), giving a vector of correlation coefficients plotted here. The vector was then checked for outliers (Grubbs’ test), which could obscure the detection of a stable change point. If there were no outliers, the residual sum of squares (RSS) between the data values and the overall mean was calculated (RSS0). Then, a candidate change point was identified as a break point that maximally reduced the RSS compared to RSS0 (best improvement). The RSS of the candidate change point (RSS1) was calculated using the data values and two segment means defined by the break (see horizontal lines). To test for significance, the observed best-improvement value (RSS1-RSS0) was compared to a permutation distribution of best-improvement values generated by randomizing cell order and recomputing the correlation matrix (n = 1000 iterations; see Methods). This approach gave a significant (P < 0.001***) change point at the 80th percentile of the separability distribution (vertical dotted line). (B) Separability at the 80th percentile of each session’s separability distribution (n = 36 control, 39 Halo sessions). The Halo group showed comparable separability values to control (LME with group as fixed effect: Estimate = -0.036, SE = 0.068, t(13.97) = -0.53, P = 0.605). (C) Same summary correlation matrix as Figure 3D but now computed from cells’ activity on milk trials (as opposed to male trials) in control sessions. (D) In control sessions (n = 36, individual lines), MALE cells showed more strongly correlated activity compared to OTHER cells (LME with MALE - OTHER coefficients as outcome variable: Estimate = 0.079, SE = 0.014, t(6.66) = 5.59, P < 0.001***). Correlation coefficients were Fisher-transformed for statistical testing. (E-F) Same as (C-D) but now in Halo sessions (n = 39). MALE and OTHER cells did not differ in their strength of correlated activity in Halo sessions (LME with MALE - OTHER coefficients as outcome variable: Estimate = 0.049, SE = 0.025, t(6.99) = 1.95, P = 0.093). (G) To detect any milk-representing cells, the lower-left triangle of the correlation matrix in (C) was taken. As in (A), each row was summarized (median, excluding diagonal), giving a vector of correlation coefficients plotted here. A break between the first and remaining values maximally reduced the RSS. However, the first value was an outlier (Grubbs’ test), which prevented considering this break as a change point. Horizontal line shows mean over all values. In (B, D, F), boxplots show median and IQR over sessions.