Introduction

A common denominator of systemic inflammatory conditions is an excessive systemic inflammatory response (SIR) [68, 78]. A promising new therapeutic field is SIR downregulation by stimulating the inflammatory reflex (IR), the intrinsic physiological neural mechanism that suppresses the production of proinflammatory cytokines [13, 16, 24, 25, 34, 57, 60,61,62, 70,71,72, 83, 84, 92]. The most widely explored IR activator is vagal nerve electrostimulation (VNS), which has been successfully used in experimental models of inflammatory disorders in animals and clinical trials in humans [7, 8, 10, 12, 14, 29, 30, 38, 40,41,42, 46, 56, 81, 87].

Recently, it has been suggested that oral NaHCO3 intake stimulates IR [65, 66]. This is highly significant because NaHCO3 is inexpensive, safe, and FDA-approved, ensuring a rapid translation into clinical practice. Oral administration of NaHCO3 solution in humans and rats induced an anti-inflammatory monocyte polarization relative to controls receiving the equivalent molar load of NaCl. Flow cytometry analysis in the spleen of NaHCO3-treated rats showed an increase in the total number of CD3 + CD4 + FoxP3 + lymphocytes (Tregs) and CD11b + CD206 + IL-10 + macrophages, whereas CD11b + F4/80 + TNF + macrophages decrease [65, 66].

The authors suggested that the immunomodulatory changes induced by oral NaHCO3 were IR-mediated through cholinergic splenic anti-inflammatory pathways (CSAP) [65, 66]. This was supported by a mitigated NaHCO3-mediated anti-inflammatory effect upon spleen displacement in association with the fibrosis of a layer of mesothelial cells attached to the spleen capsule. These cells stain positive for PGP9.5/choline esterase and lay over a dense network of nerves within the spleen capsule, thus, the authors hypothesized that the activation of mesothelial cells, by a yet unknown mechanism, releases acetylcholine near splenic nerve branches triggering CSAP. Nonetheless, spleen displacement may disrupt spleen innervation [23, 26]. Thus, it remains to be determined whether NaHCO3 can activate the IR and whether it does so via the activation of cholinergic mesothelial cells, splenic nerve branches, or both.

Here, we generated a sham (SH) and a spleen-denervated (SD) rat model that received four days ad libitum of either NaHCO3 or water (H2O). Flow cytometry analysis in the spleen replicated the NaHCO3-mediated immunomodulatory effect in SH but not in SD rats. Because both SD and SH models had disrupted connective tissue -hence mesothelial cells’ attachment to the spleen capsule- we argue for a necessary role of splenic nerve innervation to bring about NaHCO3’s effect. Furthermore, SD rats, independently from NaHCO3 treatment, modulate immunological markers.

Our study sheds light on the gut-brain-spleen communication responsible for NaHCO3’s effect on immune-cells-driven anti-inflammatory polarization. To the best of our knowledge, this is the first evidence that the immunomodulatory effect induced in the spleen by orally ingested NaHCO3 is mediated by the splenic nerve.

Methods

Rats

We used 8–12-weeks-old female (250–325 g, see Additional file 1: Fig. S1a) Sprague Dawley rats from Charles River. They were age-matched for all protocols and housed under standard conditions (12:12-h light–dark cycle and free access to food and water). All studies were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory. All experiments were conducted under the Institutional Animal Care and Use Committee Animal Use Protocol (IACUC) #20-10574. This protocol was approved by the Office of Animal Welfare, Office of Research Administration from SUNY Downstate Health Sciences University.

Surgeries and spleen collection

Rats underwent spleen denervation [26, 39, 69] confirmed by Western blotting for tyrosine hydroxylase (TH,anti-TH antibody; ab112; Abcam) as previously described [39, 85, 86]. For a description of the spleen denervation, sham surgeries, spleen collection, and immunoblotting see Additional file 1: Fig. S1a.

Rat treatments

After sham (n = 24) and spleen denervation (n = 16) surgeries, animals were kept on LS chow with ab libitum water allowing 12–14 days of recovery. On the first day of the experiment, SH and SD rats were randomly assigned to drink a vehicle (H2O, n = 18) or 0.1 M NaHCO3 (n = 22) freshly made, ab libitum (see Additional file 1: Fig. S1a for daily consumption). This treatment was kept for 4 days, animals were euthanized on day 4 and spleens were harvested for flow cytometry analysis and western blot.

The concentration for the NaHCO3’s solution (equivalent to 8.4 mg of NaHCO3 in 1 L of water) is the highest dose used in rats by Ray et al. [65, 66]. By giving 0.1 M of NaHCO3 to rats we are below non-toxic levels in humans. A safe dose of NaHCO3 in chronic kidney disease (CKD) patients is 42 mg/kg/day which is ~ 3000 to 4000 mg of NaHCO3 daily [22], this amount is equivalent to 12.6 mg in rats. Our animals consumed ~ 0.42 mg of NaHCO3 per day which is 30 times less than 12.6 mg (non-toxic NaHCO3 levels in humans). This estimate is based on the amount of water consumed ad libitum by Sprague Dowley rats between 250 and 330 g of weight, which is 50 ml [51]. NaHCO3 has a molecular mass of 84.004 g/mol, and 50 ml of NaHCO3 at 0.1 M contains 0.42 mg of NaHCO3.

Ray et al. used NaCl at 0.1 M as a vehicle instead of H2O. We did not use NaCl to avoid the potential proinflammatory confounder effect of hypertonic sodium [33, 91].

Flow cytometry

Tissue processing, antibodies, and technical information

We followed Ray et al. protocol with minor changes [65, 66]. Briefly, harvested spleens were processed and cell suspensions were incubated with antibodies and dead life solution to identify dead cells. For a full description of the flow cytometry protocol see Additional file 1: Fig. S1b, and for reagents and antibodies see Additional file 1: Fig. S2a.

To identify through flow cytometry macrophages residing in the spleen, we used antibodies against CD11bc, CD38, CD206, and TNFα. The CD38 and TNFα markers have been associated with a proinflammatory phenotype in monocytes and macrophages [31], whereas CD206 is associated with regulatory or anti-inflammatory properties [31]. The denomination of proinflammatory macrophages as M1 or anti-inflammatory as M2 oversimplifies the spectrum of phenotypes observed in these cells and there isn't a nomenclature agreed upon yet [49, 77]. For clarity, in the present work, we describe macrophages as M1-like (CD11bc + CD38 + TNFα +) and M2-like (CD11bc + CD206 +) as recently suggested [77].

Except for CD38, all the other markers have been used to identify NaHCO3-induced changes in rats [65, 66]. Ray et al. used anti-rat F4/80 antibodies for proinflammatory macrophages. At the moment of the current work, Novus had stopped the production of anti-rat F4/80 antibody and there was not a similar product available on the market. CD38 is considered selective for proinflammatory (LPS ± IFN-γ) but no anti-inflammatory macrophages (IL-4) [5], and transcriptomic analysis identified CD38 as a murine marker able to distinguish proinflammatory from anti-inflammatory-like macrophages [31]. To identify T cells, we used CD3, for T-helper we added CD4, and for Tregs we used FoxP3. We did not use CD8 but reported that CD4-T cells which include the CD8 + population.

Data acquisition

A Daily QC run was performed using Agilent Flow Cytometer QC particles to ensure the performance parameters meet the requirements. See Additional file 1: Fig. S2b for information about the flow cytometry setup. In flow experiments, the group identifiers were removed, and the analysis was performed by an investigator blind to the source of the samples. In each analysis, 500,000 total events were collected. Compensation beads (Thermo Fisher, Invitrogen™) were used to ensure that median fluorescence intensities of negative and positive were identical. We collected between 30,000 and 40000 events for compensation. Fluorescence Minus One (FMO) control was used to set the upper boundary for the background signal on the omitted label and to identify and gate the positive population. Samples were analyzed in duplicate measurements. Flow Cytometry was performed at the SUNY Downstate Health Sciences University, Flow Cytometry Facility, which received financial support from the Faculty of Medicine and grants from contributing investigators.

Data analysis

NovoExpress software was used to collect and analyze data. Dead cells and debris were excluded using forward and side scatterplots and dead life staining. Doublets were excluded with forward scatter height (FSC-H) and forward scatter area (FSC-A) plots. Representative gating images for M2-like macrophages and granularity index in SH and SD rats are shown in Fig. 1g while Tregs are shown in Fig. 2e. The granularity index in FSChigh/CD11bc cells was determined after identifying two distinct cell populations based on granularity (SSChigh/SSClow). After excluding debris and doublets we distinguish the alive populations (Additional file 1: Fig. S3a, gate R3) with lower FSC-A (y-axis) and higher FSC-A values (gate R6). We then assessed the degree of granularity (SSC values, Fig. 1B, y-axis) in both populations. Within R3 we gate agranular (SSClow) cells whereas R6 contained two populations, agranular and granular (SSChigh). The cells inside R6 (R8 and R10 gates, Fig. 1g) are CD11bc positive and are the biggest. We identify that R8 contains those more granular cells. A summary of further gating of each cell is in Additional file 1: Fig. S3.

Fig. 1
figure 1

The immunomodulatory effect induced by oral NaHCO3 on immune markers is abolished in SD animals. Spleen denervation has an independent immunomodulatory effect. SD and SH rats were randomly assigned to drink H2O or NaHCO3, spleen was processed for immunophenotyping by flow cytometry. All values represent the percentage of live cells (an expanded gating for all cells can be seen in Additional file 1: Fig. S2a), with marker positivity listed below cell subtypes. The interaction was evaluated with a two-factor ANOVA (treatment; surgery). The interaction was considered present for p < 0.15 and pse > 0.04 and ANOVA was followed by a test of single effects (Table 1). Two-Factor ANOVA results: a. CD11bc + CD38 + TNFα + cells, interaction present, p = 0.137; b CD11bc + CD206 + cells, interaction present, p = 0.006**; c CD3 + CD4 + FoxP3 + cells, interaction present, p = 0.115; d M1-like/M2-like index, interaction present, p = 0.018*; e Log10 of the Treg/M1-like index, interaction present, *p = 0.023; f SSChigh/SSClow of CD11bc + FSChigh cells, interaction present, p = 0.070. The p and pse values of interest (2-factor ANOVA, and test of simple effects) are listed on each figure if respective group effects were found important (p < 0.05 and/or pse > 0.06). Percentages of M1-like and M2-like macrophages and granularity index are indicated in corresponding quadrants. g Representative gating for M2-like macrophages and granularity index in SH and SD rats, treated with H2O or NaHCO3. FMOs control for SH and SD for M2-like macrophages is shown. The colors light red and blue represent H2O and NaCHO3, respectively. At the left of the figure is the formula used to calculate macrophages and granularity index number and the y and x-axis labels. *Significant differences p < 0.05 or moderate SE (pse > 0.06); **p < 0.01 or large SE (pse > 0.14). Figure created with NovoExpress, and Prism GraphPad

Full size image
Fig. 2
figure 2

The NaHCO3 treatment did not impact T-helper, CD11bc + CD38 + , CD4- T cells, and CD4-FoxP3 + T cells. Spleen denervation increased CD11bc + CD38 + cells and suppressed CD4-FoxP3 + T cells. SD and SH rats were randomly assigned to drink H2O or NaHCO3, spleen was processed for immunophenotyping by flow cytometry. All values represent the percentage of live cells (an expanded gating for all cells can be seen in Additional file 1: Fig. S2a, b), with marker positivity listed below cell subtypes. The interaction was considered present for p < 0.15 and pse > 0.04. Two-Factor ANOVA results: a CD11bc + CD38 + cells, no interaction p = 0.559, treatment p = 0.317, surgery p = 0.043*; b CD3 + CD4 + cells, no interaction p = 0.690, treatment p = 0.632, surgery p = 0.509; c CD3 + CD4-cells, no interaction p = 0.777, treatment p = 0.564, surgery p = 0.168; d CD3 + CD4-FoxP3 + , no interaction p = 0.861, treatment p = 0.830, surgery *p = 0.011;. The p and pse values of interest (2-factor ANOVA, and test of simple effects) are listed on each figure if respective group effects were important (p < 0.05 and/or pse > 0.06). e Representative gating for Tregs isolated from the spleen of SH and SD rats, treated with H2O or NaHCO3. The color light red represents H2O and light blue NaCHO3. To the left is the formula used to calculate Treg’s final number and y and x-axis. *Significant differences p < 0.05 or moderate ES (pse > 0.06); ** p < 0.01 or large ES (pse > 0.14). Figure created with NovoExpress and Prism GraphPad

Full size image

Cells expressing specific markers were reported as the ratio of the percentage of the events from the parent population in a specific quadrant or gate (positive population), divided by the same number plus the percentage of events from the parent population contained in the quadrant that did not express the marker (negative population). See Additional file 1: Fig. S2c for a detailed description of the proportion rate calculation for each cell type.

Statistical analysis

Data for continuous variables are expressed as means ± standard deviation (SD). We used the Kolmogorov-Smirnov test to assess the normal distribution and we applied log10 transformation to those non-parametric variables (see Table 1). Logarithmic transformation was applied to variables with a non-parametric distribution resulting in a parametric distribution. A two-factor ANOVA was used to compare the four groups (SH surgery drinking H2O or NaHCO3, or SD drinking H2O or NaHCO3) and assess for interaction between treatment (H2O or NaHCO3) and surgery (SH, or SD). The test of single effects was used to discriminate the difference between groups if an interaction was present [32], Support, 2022 #306). One-way ANOVA was used to compare weight at euthanasia and amount of H2O or NaHCO3 solution consumption. The Levine test assessed for equal variances (see Table 1). Multivariate analysis was done using Artificial Neural Networks (ANN) and principal component analysis (PCA). We did not assess collinearity in the ANN models because of good predictive capacity in presence of collinearity [88]. We assessed multicollinearity before PCA excluding variables with a variance inflation factor (VIF) above 5 [88].

Table 1 Descriptive statistics, two-factor ANOVA, and test of simple effects analysis for immune markers after immunophenotyping with flow cytometry
Full size table

To assess the magnitude of the difference between variables we used the effect size (ES) and the probability. The ES was measured by the partial eta squared (pes) and it was labeled “important” when pes was above 0.06 (moderate = 0.14 > pes > 0.06; high = pes > 0.14) [35]. The significance level was defined as p < 0.05. We provide both measurements in tables and Figures. All the graphs and data tables were designed using Prism GraphPad, and the statistical analysis was performed with SPSSv28 software.

Results

Effect of oral NaHCO3 intake in SH and SD animals

SD and SH rats were randomly assigned to drink a vehicle (H2O), or a solution of NaHCO3 (0.1 M) ad libitum for 4 continuous days. After spleen processing, immune cell phenotypes were identified by flow cytometry (see Additional file 1: Fig. S3b). Four groups of animals were stratified in 2 factors, surgery (SH or SD) and treatment (H2O or NaHCO3). The means and standard deviations for all variables are shown in Table 1. To control a possible interaction between spleen denervation and treatment response to oral NaHCO3, we used a two-factor ANOVA.

The variables with interaction (Table 1; p < 0.15, pes > 0.04) included CD11bc + CD38 + TNFα + cells (M1-like macrophages), CD11bc + CD206 + cells (M2-like macrophages), log10 of the M1/M2 ratio (M1-like/M2-like ratio), SSChigh/SSClow ratio in CD11bc + FSChigh cells (granularity index in CD11bc/FSChigh cells), CD3 + CD4 + FoxP3 + lymphocytes (Tregs), and the ratio of Tregs/M1-like macrophages (Tregs/M1).

Within this group, oral NaHCO3 decreased the fraction of M1-like macrophages and M1-like/M2-like ratio in SH animals compared to the H2O-treated group (Fig. 1a, d; stats), crucially, this effect was not observed in NaHCO3 and H2O treated-SD animals (Fig. 1s; stats). The granularity index of CD11bc/FSChigh also decreases in the NaHCO3-SH group (Fig. 1f; stats), while remaining unchanged in the SD animals (Fig. 1f; stats). This variable decreases in the same direction as proinflammatory monocytes, this is supported by a positive correlation with M1-like and negative with M2-like macrophages (Pearson coefficient 0.723, p < 0.01; − 0.522, p < 0.01 respectively) without collinearity with either macrophage subtype (Additional file 1: Fig. S5b). In addition, NaHCO3 treatment increased M2-like macrophages, Tregs, and Tregs/M1-like ratio in SH animals compared to the H2O group (Fig. 1b, c, e; stats); and like the M1-like findings, the NaHCO3-induced rise in M2-like, Treg, and Treg/M1-like was not observed in SD animals (Fig. 1b, c, e; stats).

The variables without interaction (Table 1; p > 0.15, pes < 0.04) between treatment and surgery involve CD11bc + CD38 + cells (CD11bc + CD38 + cells; Fig. 2A), CD3 + CD4 + lymphocytes (T-helper; Fig. 2B), CD3 + but CD4 negative lymphocytes (CD4- T cells; Fig. 2c), and lastly CD3 + FoxP3 + and CD4 negative lymphocytes (CD4-FoxP3 + T cells, Fig. 2d). None of these variables differed between the groups in response to treatment. Moreover, the T-helper and CD4-T cells were similar across the SH and SD animals for both H2O and NaHCO3 groups.

In summary, in the SH animals, treatment with oral NaHCO3 in comparison with H2O increased M2-like macrophages, Tregs, and Tregs/M1-like ratio while decreased M1-like macrophages, the granularity index of CD11bc/FSChigh, and M1-like/M2-like ratio. Other cells were not affected including T-helper, CD11bc + CD38 + cells, CD4-T cells, and CD4-FoxP3 + T cells. In SD animals, the comparison of these immunological markers between H2O and NaHCO3 remained largely unchanged (Table 1, Figs. 2, and 3). Lastly, the interaction between surgery and treatment suggests an independent immunomodulatory effect induced by spleen denervation (Table 1).

Fig. 3
figure 3

Measurement tools for all the ANN models. a Curves built with precision, recall, accuracy, and F1 score values for the training set are consistently above 80% for SH in comparison with the SD group. b Curves built with precision, recall, accuracy, and F1 score values for the testing set are consistently higher for the SH in comparison with the SD group. Figure created with Prism GraphPad

Full size image

Effect of spleen denervation on immune markers.

This immunomodulatory effect of spleen denervation can be better seen by looking at the test of simple effects between SH and SD animals in each treatment group (Table 1). Within the H2O group, spleen denervation (compared to H2O/SH) moderately (0.14 > pes > 0.06) decreases M1-like macrophages, Tregs, and the granularity index of CD11bc/FSChigh cells, and did not change M2-like macrophages (Table 1). Within the NaHCO3 group, there were no changes (p > 0.05, and/or pes < 0.06) between SD and SH animals for M1-like macrophages or the granularity index. Contrary, SD rats have fewer M2-like macrophages and Tregs (p < 0.05, pes > 0.14, Table 1). These findings point to an effect of spleen denervation on immune markers in both H2O and NaHCO3-treated animals.

In the absence of an interaction between treatment and surgery (where the simple effect test is unnecessary), spleen denervation globally decreased CD4-FoxP3 + T cells and increased CD11bc + CD38 + cells in comparison to SH animals (Table 1). Other cells were unaffected by the surgery such as T-helper, and CD4-T cells. Overall, these analyses further indicate that spleen denervation, independently of the treatment effect (NaHCO3 or H2O), has an immunomodulatory effect.

Identifying the multivariate effect of spleen denervation by artificial neural network (ANN) classification

Our findings are consistent with the immunomodulatory NaHCO3-mediated effects described by Ray et al. [65]. Crucially, our data shows the abolishment of this effect by spleen denervation. Next, we assess the same question with a multivariate approach. We assume that if the effect of NaHCO3 is mediated by the splenic nerve, a classificatory model integrated by multiple immune markers would generate accurate predictions in SH but not in SD animals. To build the model we used ANN [2, 27, 37, 43, 64]. We included those immune markers with an important difference (p < 0.05 and/or pes > 0.06) between NaHCO3 and H2O groups in the ANOVA results (Table 1).

The SH group has 11 and 13 animals subjected to H2O and NaHCO3 respectively, whereas the SD group has 7 and 9. The models were generated with SPSS running a simultaneous and parallel analysis for SH and SD groups. We ran five multilayer perceptron (MLP) ANN models (ANN1–ANN5) per group (Additional file 1: Fig. S4b, c). 70% of the sample was used to train the ANN models and 30% to test them which makes a rate of 2.3 (70/30 = 2.3). All models contained 6 independent variables, and the outcome-dependent variable was treatment (H2O or NaHCO3). To explore the overall performance of the ANN models (5 per group) we examined precision, recall, accuracy, and F1-score (HN, 2019).

All ANN models generated in SH performed consistently better than the models in SD animals (Fig. 3a, b). We selected ANN5 (ANN5-SH, and ANN5-SD) because it has the most comparable training and testing partition rates between SH and SD groups and is the closest to a 2.3 ratio (Additional file 1: Fig. S4a). Hence, a sample of 16 SH and 10 SD animals (66.7% and 62.5% respectively) randomly assigned by SPSS was used as the training, leaving aside 8 SH (33.3%) and 6 SD (37.5%) rats to validate ANN5-SH and ANN5-SD respectively (see network information in Additional file 1: Fig. S4b).

Figure 4a, b display the predictive pseudo-probability for ANN5-SH and ANN5-SD. Within the H2O category, from left to right, the ANN5-SH model correctly classified more animals as drinking water (Fig. 4a; red boxplot in H2O category) than incorrectly classified them as drinking NaHCO3 (Fig. 4a; blue boxplot in H2O category). Consistently, within the NaHCO3 category, the ANN5-SH model correctly classified animals -with a predicted probability close to 1- as drinking NaHCO3 (Fig. 4a; blue boxplot in NaHCO3 category), and marginally incorrectly classified them -with a predicted probability close to zero- as drinking water (Fig. 4a; red boxplot in NaHCO3 category). In contrast, the ANN5-SD model showed a 0.6 probability of being classified as NaHCO3 and a 0.4 probability of being classified as H2O for either category, thus failing to correctly classify animals to each category (Fig. 4b).

Fig. 4
figure 4

Predictive pseudoprobability, ROC, and variables’ normalized importance for ANN5-SH and ANN5-SD models. a The ANN5-SH’s predictive pseudo-probability to correctly classify rats drinking H2O or NaHCO3 is close to 100%, whereas the misdiagnosed cases were close to zero. b The ANN5-SD’s predictive pseudo-probability for correct predictions was close to 0.6 (60%) whereas the misdiagnosed cases were at 40%. c ANN5-SH’s ROC curve, representing the sensitivity (y-axis) and specificity (x-axis) for the model, is supported by an excellent AUC of 0.972. d ANN5-SD’s ROC curve and an unsatisfactory AUC of 0.587. e, f The importance of the 6 independent variables used to build the ANN5 models, e corresponds with ANN5-SH and f with ANN5-SD. Figure created with SPSS 28 and Prism GraphPad

Full size image

The classificatory robustness of ANN5-SH and ANN5-SD was further compared by using the confusion matrix of cases classification, precision, recall, accuracy, F1 score, and area under the Receiver Operating Characteristics (ROC) curve (AUC), (HN, 2019). A case classification as a confusion matrix for both models is shown in Table 2. The outcome of the ANN models was defined as correct if the predicted probability was above 0.5. ANN5-SH correctly classified 15 out of 16 cases (93.8%) in the training and 7 out of 8 cases (87.5%) in the testing data sample (Table 2). In contrast, ANN5-SD correctly classified 6 out of 10 cases (60%) in the training set and 3 out of 6 cases (50%) in the testing set (Table 2). Thus, the ANN5-SH model performed substantially better than the ANN5-SD to classify treatment exposure, which is consistent with our prediction.

Table 2 Classification is reported as a confusion matrix
Full size table

Precision, recall, accuracy, and F1 score for ANN5-SH are above 90% in the training and above 80% in the testing set (Table 3). The F1 score, is above 90% in the training set and 88.9% in the testing set, which is considered very good and good respectively [74]. Contrary, the F1 score for ANN5-SD was lower at 75% and 66.7% (Table 3). Furthermore, ANN5-SH had an accuracy of 93.8% and 87.5% for ANN5-SH, over the 70% considered as good performance in machine learning [93]. In contrast, the accuracy for ANN5-SD was 60% for the training and 50% for the testing set (Table 3). The recall was 100% in the training and testing sets for both groups (Table 3).

Table 3 Precision, recall, accuracy, and F1 score for ANN-SH and ANN-SD animals’ models
Full size table

The ANN5-SH and ANN5-SD models were further validated by examining the area under the Receiver Operating Characteristics (ROC) curve (AUC), which displays the relationship between true and false positive rates (HN, 2019). The prominent left-shift (from the diagonal representing equal true positive/false positive rate) of the ROC curves in the ANN5-SH model indicates high sensitivity and specificity, and more accurate classification for both the H2O and NaHCO3 categories (Fig. 3e). The AUC value, which estimates the quality of the classificatory model, for the ANN5-SH model was 0.972, which is considered excellent [9, 54]). In contrast, the ROC curves for either category in the ANN5-SD model remained unshifted (from the diagonal, Fig. 3f), and generated an AUC value of 0.587, which is unsatisfactory [9, 54].

Lastly, the importance of the 6 independent variables used to build the ANN5 models was ranked from high to low, those that contributed the most to ANN5-SH were the granularity index of CD11bc/FSChigh (100%), Tregs (79.1%), and M2-like macrophages (58.4%) (Fig. 4c). The variables that contributed most to the ANN5-SD model were the M1-like/M2-like ratio (100%), the granularity index of CD11bc/FSChigh (88.4%), and Tregs/M1-like ratio (48.5%) (Fig. 4d).

In summary, ANN models built on information from the SH group can correctly classify whether animals were drinking H2O or NaHCO3, but one built on the same information from the SD group failed to do so, which is consistent with a splenic nerve role in NaHCO3-mediated immunoregulation.

Identifying immune markers clusters underlying immunoregulation by principal components analysis (PCA)

In “Effect of oral NaHCO3 intake in SH and SD animals” and "Effect of spleen denervation on immune markers" we identified interactions between treatment and surgery on several immune markers. We, therefore, wondered whether the immunoregulatory effects of spleen denervation and treatment are differentiated. We use PCA to address this question.

We first ran PCA with the 12 variables including 10 immune markers plus the type of treatment and surgery to examine the correlation matrix, collinearity, and anti-image matrices. Eleven of the 12 variables correlated above 0.3 (Additional file 1: Fig. S5a), 4 variables showed high M1-like multicollinearity coefficients (VIF > 5 and tolerance < 0.2; Additional file 1: Fig. S5b) and one variable (CD11bc + CD38 + cells) had an anti-image correlation matrix below 0.50 (Additional file 1: Fig. S5c). Based on this preliminary assessment, we excluded M1-like/M2-like ratio, Treg/M1-like ratio, T-helper, and CD4-T cells and CD11bc + CD38 + cells, keeping 7 variables for further analysis.

A new PCA with the 7 variables shows a Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy of 0.64 (above the recommended value of 0.60) and a significant Bartlett’s test of sphericity (χ2 (21) = 59.607, p < 0.001). Screen plot and parallel test analysis [58, 59] identified two components with eigenvalues over 1.134 (Fig. 5a). The first and second components’ eigenvalues explained 33% and 25.5% of the variance respectively, producing a cumulative variance of 58.43% (Table 4). The commonalities are above 0.30 supporting that each variable shares common variance with others (Table 4).

Fig. 5
figure 5

Principal component analysis (PCA) for SH and SD groups together. a Scree plot from PCA showing as many components as variables (x-axis), the parallel analysis for the eigenvalue (y-axis) for component 3 was 1.134 and a dashed line is at that level including only the first two components in the final PCA. b The components plot in rotated space including component 1 as an orange area (immunoregulatory effect of spleen denervation) and component 2 as a blue area (immunoregulatory effect of NaHCO3 treatment). The first component is loaded with Treg, CD4-FoxP3 + T cells, M2-like macrophages, and surgery (zero representing sham surgery and one spleen denervation). The second component is loaded with M1-like macrophages, the granularity index, and the treatment (zero representing H2O and one NaHCO3 treatment). The area where blue and orange overlap represents the only complex variable in the PCA, which is M2-like macrophages. These cells had a predominant loading factor of 0.518 on Component 1, however, the loading for Component 2 was close to 0.427. Figure created with SPSS 28, and Prism GraphPad

Full size image
Table 4 The loading factors, communalities, variance, and eigenvalues of PCA combining SH and SD animals in the same group
Full size table

An assessment of the components’ structure showed that an orthogonal rotation and quartimax offers the best-defined structure (Fig. 5b). The first component is loaded with Tregs, CD4-FoxP3 + T cells, and M2-like macrophages, in inverse correlation with the presence of spleen denervation (Table 4; orange rectangle in Fig. 5b). The second component is loaded with M1-like macrophages and the granularity index of CD11bc/FSChigh cells, both variables in negative correlation with the presence of NaHCO3 treatment (Table 4; blue rectangle in Fig. 5b). Notably, M2-like macrophages had a cross-loading correlation in both components (Table 4). The PCA’s composite reliability to assess internal consistency reliability [21, 79] is 0.837 and 0.769 for components one (4 items) and two (3 items), respectively, which are above 0.7 and considered adequate [1].

Because spleen denervation and NaHCO3 treatment modulate M2-like macrophages and Tregs, we conduct a PCA only in SH animals. We included the same variables used in the PCA analysis for both groups. For the SH group, the variable CD4-FoxP3 + T cells had an anti-image correlation matrix below 0.50 and was excluded. The KMO measure of sampling adequacy was 0.73, and Bartlett’s test of sphericity was significant (χ2 (10) = 43.750, p < 0.001). The parallel test identified a first component that explained 57.17% of the cumulative variance and had a composite reliability of 0.866 (Table 5). The second identified component minimally contributed to the cumulative variance (Eigenvalue cut-off of 1.233); thus, we retained only the first component. This unique component was loaded (> 0.45 loading factor) with the presence of NaHCO3 treatment, M2-like macrophages, and Treg, in inverse correlation with M1-like macrophages, and the granularity index (Table 5). In contrast, for the SD group, the KMO measure of sampling adequacy was 0.56 (below the recommended value of 0.60) and Bartlett’s test of sphericity was not significant (χ2 (10) = 6.960, p = 0.729), thus, we did not carry out further analysis.

Table 5 Loading factors, communalities, variance, and eigenvalues for PCA for the SH group alone
Full size table

Overall, PCA analysis for all animals (SH and SD) treated with NaHCO3 or H2O distinguished two distinct clusters of factors underlying immunomodulation, and PCA analysis within the SH group produced a sole component. These analyses are supportive of a multivariate immunomodulatory effect of oral NaHCO3 treatment.

Discussion

Ray et al. recently demonstrate a decrease in M1-like, and an increase in M2-like macrophages, and Tregs, in rats receiving 0.1-M NaHCO3 solution for three consecutive days [65]. The authors hypothesize IR activation as the possible mechanism responsible for this effect. Herein, we replicated the NaHCO3-mediated immunomodulatory effect described by Ray et al. in SH but not in SD animals. Thus, suggesting the necessity of the splenic nerve to bring about NaHCO3’s effect. Furthermore, SD, independently from NaHCO3 treatment, modulated some immune cells whereas others were unaffected. Taken together, these data shed light on the gut-brain-spleen communication responsible for NaHCO3’s effect on immune-cells-driven anti-inflammatory polarization. In addition to describing a sole immunomodulatory-spleen denervation effect on specific immune cells. To the best of our knowledge, this is the first evidence that the immunomodulatory effect induced in the spleen by orally ingested NaHCO3, is mediated by the splenic nerve.

Immunomodulatory effect of oral NaHCO3 intake in SH animals

Ray et al. hypothesize that the effect of NaHCO3 in macrophage polarization was by IR activation [65, 66]. In the classic CSAP model [83], upon IR stimulation, CSAP converging into the splenic nerve releases norepinephrine (NE) which activates a subset of T cells in the spleen through β2-adrenergic receptors (β2AR). These T-cells release acetylcholine (ACh) which interact with α7 nicotinic acetylcholine receptors (α7nAChR) in spleen macrophages downregulating the production of proinflammatory cytokines but without increasing IL-10 production [70]. The IR can also be stimulated by splanchnic sympathetic pathways (SAP) which also feed the splenic nerve, increasing IL-10 production [48]. In parallel, NE can also bind to β2AR on macrophage s to induce an anti-inflammatory polarization [26], as well as to β2AR on Tregs, proliferation, and enhancing Treg function [19]. Furthermore, the activation of β2-receptors on splenic regulatory lymphocytes via splenic nerve-norepinephrine released upon vagal nerve stimulation suppresses LPS-induced inflammation, providing additional evidence for the anti-inflammatory properties of these cells [89].

In our study, oral NaHCO3 modulates M1-like macrophages, M2-like macrophages, Tregss, the ratios of M1-like/M2-like, Tregs/M1-like, and the granularity of CD11bc/FSChigh cells. Other cells were not affected including T-helper, CD11bc + CD38 + cells, CD4-T cells, and CD4-FoxP3 + T cells. These results suggest that NaHCO3 activates the IR.

Notably, we find that among all immunological variables quantified, the largest changes between H2O and NaHCO3 groups are observed in Tregs/M1-like and M1-like/M2-like ratios. We anticipate these results based on the single response of M1-like macrophages (suppression), M2-like macrophages (increase), and Tregs (increase) to oral NaHCO3 (Table 1, Figs. 1, 2). However, the expression of these immune markers as a ratio displays a bigger effect size (pes = 0.186, pes = 0.174,) and significance (p = 0.007, p = 0.009) between the treatment groups. Similarly, Ray et al. showed that the monocyte polarization was clearest when expressed as the number of M1-like/M2-like [65, 66]. In the same line, we surmise that the more important difference of Tregs/M1-like and M1-like/M2-like ratios in comparison with M1-like or Treg alone speaks of a functional link between monocytes and Tregs. Macrophages with a more anti-inflammatory phenotype and Tregs seem to regulate each other and generate an immunosuppressive loop through cytokines or cell-to-cell interaction [17, 45, 67, 75, 80, 82].

We found a novel decrease in the granularity index of CD11bc + /FSChigh cells in SH animals treated with NaHCO3 in comparison with the H2O group. We distinguished two cell populations based on granularity: agranular (SSClow) and granular (SSChigh) and identified CD11bc-positive cells in the latter population, which were the biggest. Previously, a higher proportion of granular than agranular cells has been reported in tumoral cells in an inflammatory environment [47]. The granularity index correlates positively with M1-like and negatively with M2-like macrophages but without collinearity with either macrophage subtype. This suggests that changes in the granularity index of CD11bc + /FSChigh cells move in the same direction as M1-like cells but relying on a different type of cell (lack of collinearity), a potential suggestion is neutrophils. Consistent with this, Ray et al. described a significant decrease in neutrophils (CD16 + TNFα +) in the blood of human subjects after a load of oral NaHCO3 [65].

In addition to finding a sole effect under oral NaHCO3 intake, where the number of M2-like macrophages and Tregs increase while M1-like and the granularity index of CD11bc + /FSChigh cells decreases, we also find a multivariate association among these variables with treatment. In SH animals, PCA clusters in one component NaHCO3 treatment, Tregs, and M2-like macrophages in negative correlation with M1-like macrophages and granularity index of CD11bc + /FSChigh cells. We name this component immunomodulatory effect of NaHCO3 and suggests an orchestrating immunomodulatory role for NaHCO3 treatment. As discussed above, the increase in M2-like macrophages increments Tregs, which can further increase M2-like cells. This coordinating effect seems consistent with the notion that NaHCO3 triggers changes that match the effect of known IR activators [4, 76]. Accordingly, VNS, an established IR activator, reduces neutrophil migration [36], promotes microglial M2-like polarization [15], and reduces lung mRNA levels of M1-like macrophage markers, while increasing M2-like markers [44]. Furthermore, nicotine, another IR activator, reduces neutrophil recruitment during sepsis development [36], and induces both anti-inflammatory macrophage polarization [73] and Tregs number [52].

Splenic nerve and oral NaHCO3 intake

NaHCO3-mediated immunomodulation of immunological markers bespeaks involvement of the IR pathway, but a critical test was to demonstrate the necessity of the splenic nerve to NaHCO3’s effect. SD animals that received NaHCO3 treatment did not show changes in M2-like macrophages, M1-like macrophages, granularity index of CD11bc + /FSChigh cells, M1-like/M2-like ratio, Tregs, or Tregs/M1-like ratio. Moreover, ANN analysis correctly classifies whether animals were drinking H2O or NaHCO3 only for the SH group but fails to do so for the SD group. As above, a PCA in the SH group gathers M1-like macrophages, granularity index of CD11bc + /FSChigh cells, M2-like macrophages, Tregs, and NaHCO3-treatment in one component. However, an attempt to run PCA analysis in SD animals using the same variables fails.

These results demonstrate that the splenic nerve is necessary to observe oral NaHCO3-mediated immunomodulatory changes in the spleen. This effect seems independent from the mesothelial cells’ choline esterase + described by Ray et al. [65, 66]. We think so because moving the spleen to the midline of the abdominal cavity was not sufficient to disrupt the immunomodulatory shift seen in SH animals. The splenic nerve is a necessary part of the IR to suppress systemic inflammation. This has been tested in the context of endotoxemia for CSAP and SAP, the efferent arms of the IR. Ballinas-Rosas et al. demonstrated how the splenic nerve was required for CSAP control of TNFα during endotoxemia [69]. Similarly, Martelli et al. report that splenic nerve activity is dependent on inputs from sympathetic splanchnic nerves during endotoxemia, while is absent when splanchnic nerves have been severed. Moreover, 80% suppression of TNFα is only present in those animals with intact splanchnic nerves [48]. By denervating the spleen, we have disrupted SAP and CSAP as they converge in the splenic nerve. Thus, this finding brings strong support to classify NaHCO3 as an IR activator.

We have previously suggested a mechanistic model whereby oral NaHCO3 intake stimulates the IR, in which the immunomodulatory effects of NaHCO3 might be mediated by choline esterase + mesothelial cells attached to the spleen capsule as well as splenic nerve branches [4]. By stimulating vagal afferents NaHCO3 oral intake could activate the nucleus of the tractus solitaries and other IR-implicated brain regions, the output of which could trigger the IR via sympathetic (SAP) and parasympathetic (CSAP) pathways converging in the splenic nerve. Additionally, via impacting afferent vagus, NaCHO3 might stimulate other vagal anti-inflammatory pathways that are not dependent on the spleen [11, 50]. Further experimental assessment within the brain and in animal models of LPS-induced endotoxemia and systemic inflammatory conditions, with disruption of SAP or CSAP, is crucial to validate and refine the characterization of NaHCO3's properties as an inflammatory reflex (IR) activator.

Due to successful outcomes in prior studies with inflammatory reflex (IR) activators, NaHCO3 holds promise for treating various inflammatory conditions, including chronic kidney disease, atherosclerosis, hypertension, coronary artery disease, stroke, cancer, diabetes mellitus type 2 (DM2), obesity, Alzheimer's disease, autoimmune diseases, and psychiatric and neurological disorders[28], Matei, 2022 #618, Annoni, 2019 #619, Li, 2022 #616, Reijmen, 2018 #620, Sorski, 2023 #621, Dai, 2020 #622, Vargas-Caballero, 2022 #623, Ng, 2020 #624, Cimpianu, 2017 #625,[90]. Furthermore, NaHCO3, commonly used for heartburn and metabolic acidosis, has mild side effects [18] and its role as an IR activator suggests a straightforward translation to clinical use, leveraging its FDA approval and cost-effectiveness. We have recently published a review article that extensively addresses the potential clinical relevance of NaHCO3 for treating various inflammatory conditions [4].

Spleen denervation has an independent immunomodulatory effect

We found that spleen denervation-independent of H2O or NaHCO3 treatment- decreased M1-like macrophages, granularity index in CD11bc/FSChigh cells, Tregs, M2-like macrophages, and CD4-FoxP3 + T cells whereas T-helper and CD4-T cells remained unaffected. The effect induced by SD on proinflammatory markers (M1-like, and granularity index CD11bc + /FSChigh cells) was more notable when comparing SH-H2O with SD-H2O groups, which was expected as these makers decrease with NaHCO3 treatment. Contrary, NaHCO3’s enhancing effect on anti-inflammatory immune markers (M2-like macrophages, and Tregs) sets up a more evident difference between SH and SD animals in the NaHCO3 group (Table 1). The immunomodulatory SD effect was further evidenced by a PCA component that grouped Tregs, CD4-FoxP3 + T cells, and M2-like macrophages inversely correlated with the presence of spleen denervation.

Consistent with a lower Treg percent in our SD group, an increase in Tregs observed in mice exposed to repeated social defeat stress (RSDS) was abolished in SD animals, which also showed decreased levels of IL-2, IL-17A, and IL-22 (cytokines specific from T cells) whereas T helper cells, IL-6, TNF-α, and IL-10 were unchanged [20]. We also found decreased CD4-FoxP3 + T cells with SD which includes CD8 + FoxP3 + T cells [53]. The function of CD8 + FoxP3 + T cells is not well characterized; however, suppressive properties of CD8 + FoxP3 + T cells in vitro and in vivo have been reported [55], also transcriptomic analysis in murine-induced CD8 + FoxP3 + T cells resembles CD4 + Tregs [3]. Thus, spleen denervation impacts specialized T cells (Tregs, CD4-FoxP3 + T-cells) whereas T-helper and CD4-T cells remain unchanged.

CD11bc + CD38 + cells were increased by spleen denervation. CD38 is present in multiple cell types but is most abundant in hematopoietic cells [63]. In the immune cells, CD38 is expressed in B cells, DC, NK cells, T cells, monocytes, macrophages, and neutrophils among others [63]. This molecule can act as a receptor for CD31 and as a NAD-depleting enzyme having an important role during inflammation [31, 63]. Thus, CD11bc + CD38 + cells in our study may mostly include monocytes, macrophages, DCs, and possibly some NK, B, and T cells, which total number increase by spleen denervation.

A study from Kooijman et al. showed that selective parasympathetic spleen denervation (by clearing connective tissue in the poles of the spleen) increased the count of splenic DC, B cells, and T cells, and gene expression of proinflammatory cytokines in the liver and peritoneal leukocytes in comparison with SH [39]. Additional sympathetic spleen denervation (by clearing of connective tissue in the hilum and splenic arteries) increased circulating IL-1β and IL-6 [39]. Our surgical spleen denervation combined the two procedures completed by Kooijman et al. We did not measure cytokines, but except for an increase in CD11bc + CD38 + cells (potentially representing monocytes, macrophages, and DCs), the effect in other immune markers suggests a predominant immunosuppressive effect of spleen denervation.

Our data support an independent immunomodulatory effect of SD: it decreased more specialized and polarized cell types such as Tregs, CD4-FoxP3 + T cells, M2-like macrophages; it increased CD11bc + CD38 + cells whereas CD4-T cells and T-helper cells were unaffected. More complete phenotypic panels are needed to better characterize the impact of spleen denervation on these cell populations.

Limitations

Our study has implications but is not without limitations. We did not carry full immunophenotyping for some of the cells reported in the study such as the granularity index of CD11bc + /FSChigh cells, CD11bc + CD38 + cells, and CD4-FoxP3 + T cells which limits and makes the interpretation speculative. We worked with female rats as this sex is predominantly affected by autoimmune and some other inflammatory conditions [6], future work needs to expand our results to males. Also, the multivariate analysis specifically PCA would have benefited from a bigger sample size to conduct confirmatory factor analysis. Despite these limitations, this study contributes to supporting the role of NaHCO3 as an IR activator which opens a wide broad spectrum of therapeutic possibilities.

Conclusions

We report that the splenic immunomodulatory changes induced by oral NaHCO3 in SH are abolished in SD animals. This immunomodulatory effect is multivariate, where immune markers are grouped in the same component but proinflammatory cells are polarized on one side, inversely correlating with anti-inflammatory markers and NaHCO3 treatment located on the opposite side. Thus, further supporting an orchestrating immunoregulatory effect of oral NaHCO3. To our knowledge, this is the first evidence that the splenic nerve plays a necessary role in communicating signals induced from an oral basic solution to the spleen. This evidence may be enough to classify oral NaHCO3 as an IR activator. We also report an impact of spleen denervation on splenic immune markers, independent from treatment. This effect was immunosuppressive on more specialized T cells whereas other less specialized ones did not change.