The intended lesions and reconstructed ACC damage based on histological evaluation are shown in one hemisphere for four animals in Figure 2A and B, respectively. The ACC lesion covered most of the target cytoarchitectonic areas 24a and 24b of Paxinos et al., 2011, just above the corpus callosum. The rostral limit of the lesions was adjacent to the genu of the corpus callosum, and the caudal limit was just anterior to area 23a caudally. Dorsally, the lesions extended past 24b into motor area 6M. There was little, if any, encroachment into subgenual area 25. Apart from one case which showed some sparing of the lesion in the left hemisphere, there was extensive overlap in the placement of the ACC lesion. The anterior-posterior extent of the bilateral lesion for each monkey is compiled in Figure 1—figure supplement 1 as sections from MRI images.

Representative photomicrographs of the ACC lesion and a control are presented in Figure 2B, which shows the distribution of myelinated fibers in the ACC region stained using a high-resolution Black-Gold II myelin stain (Histo-Chem Inc, Jefferson, AR, USA). There was clear evidence that the lesion created a major disruption to the normal radial arrangement of the fibers in the ACC region caused by extensive demyelination of the axons (Figure 2B). The ACC lesion also impacted the integrity of white matter tracts local to the site of the lesion (Figure 2—figure supplement 1), but the transverse diameter of major fiber tracts, namely the corpus callosum and the anterior commissure, did not differ between the groups. The loss of neurons, however, and the respective increase in glial cells at the lesioned site, especially at the interface between the gray and white matter, was clearly observed (Figure 2C and D).

We examined the cellular and neurotransmitter composition of the AMY and PAG, as these structures are downstream from the ACC and their natural development may be affected by the infant ACC lesions. We first investigated whether neurons in these structures were degenerated using Fluoro-Jade C (Histo-Chem Inc), which is used as a marker for apoptotic, necrotic, and autophagic cells. There was no sign of neurodegeneration in these downstream brain regions 2 years following the infant lesion. We next examined the proportion of neurons in the AMY and PAG expressing GABA, since changes in the relative number of inhibitory interneurons could serve as a marker for downstream neuroplasticity in response to the ACC lesion (Atapour et al., 2021). We found a significant reduction of GABA positive neurons in two structures, the basomedial AMY and the dorsal portion of the PAG (Figure 2E). This reduction suggests a disruption of the normal inhibitory balance within the vocal network following the infant ACC lesions.

In the weeks following bilateral ACC lesions, infants tested in the isolated chamber continued to vocalize readily, which is somewhat surprising given the critical role of this structure in normal vocal behavior (Jürgens and Ploog, 1970; Müller-Preuss et al., 1980; Jürgens, 1976). From (presurgical) postnatal week 2 to (postsurgical) postnatal week 6, we annotated 23,000 calls from the five lesioned and five control marmosets. Sample spectrograms from audio recordings of a twin pair before and after surgery are shown in Figure 3A and B. While there was variability among individuals, the calls were complex and diversified from postnatal week 2, consisting of cries, as well as immature versions of adult vocalizations, including phee, twitter, and trills, as well as complex calls when two calls merged such as trill-twitter or cry-phee, or any other combination (Figure 3A). By the sixth postnatal week, the call repertoire for both the lesioned animals and the controls had both evolved, with no conspicuous difference between groups (Figure 3B). Most notably, the relative reduction of the total rate and diversity of calls was similar between groups (Figure 3C, χ²(2)=2.8464, p=0.24), with the reduction matching the known maturational changes accompanying growth of the vocal apparatus and increased respiratory powers (Zhang and Ghazanfar, 2018). Further analysis, on the proportion of calls emitted following surgery, showed that the ACC lesion had minimal effects on the rate of most call types during this period: phee (β=–0.07, 95% CI [–0.31, 0.17], p=0.49); twitter (β=–0.07, 95% CI [–0.16, 0.01], p=0.09); trill (β=–0.03, 95% CI [–0.11, 0.06], p=0.49); cry (β=0.13, 95% CI [–0.03, 0.29], p=0.10) (Figure 3D). The exception to this rule was an elevation in the rate of ‘other’ calls, which comprised tsik, egg, ock, chatter, and seep calls. These calls were significantly elevated in animals after the ACC lesion (β=0.11, 95% CI [0.03, 0.20], p=0.018). This was driven mostly by an increase in the lesion group during postnatal week 4.

Two additional variables relatively unaffected by the ACC lesion were the call durations and inter-call intervals, acoustic features that have been used to track vocal development in previous studies (Takahashi et al., 2015; Gultekin and Hage, 2018; Gultekin et al., 2021). Consistent with the overall decrease in vocalization rate with increasing age, there was an associated increase in inter-call intervals which was noted at late postnatal weeks (β=0.33, 95% CI [0.22, 0.43], p<0.001) which held true for both controls and lesioned group (χ²(2)=1.88, p=0.39). None of the specific call types exhibited developmental changes in call type duration phee (χ²(2)=1.08, p=0.58); trill (χ²(2)=2.87, p=0.24); twitter (χ²(2)=2.79, p=0.25); cry (χ²(2)=0.057, p=0.97), but there were slight changes in the duration of phee syllables exhibited only by animals with an ACC lesion, which we discuss later. Importantly, we did not observe ACC-lesion induced changes in physical growth factors such as body weight and grip strength which could feasibly impact vocal parameters such as duration (Takahashi et al., 2015; Figure 3—figure supplement 1).

Whereas many of the basic vocal parameters evolved normally in the animals with the ACC lesions, one major difference related to their use of social contact calls. By 6 weeks of age, marmoset vocalizations are known to approach their mature state and become dominated by social contact calls, namely phees, trills, and twitters. The phee call is studied most extensively as a long-distance contact call. It is typically evoked when the animal is socially distanced or isolated, and it promotes vocal exchanges between marmosets located out of sight in far-away locations (Ghazanfar et al., 2001; Miller and Wang, 2006) to facilitate reunion with family groups (Bakker and Langermans, 2018). Trills and twitters are short-distance contact calls thought to monitor the presence of group members (Bezerra and Souto, 2008; Liao et al., 2018). Since the ACC plays a major role in socioemotional cognition (for review, see Devinsky et al., 1995), we surmised that the ACC lesion might specifically influence the socioaffective content of the vocalization that is normally expressed through contact calls.

We thus grouped phee, twitter, and trill calls as mature contact calls and compared them with noncontact calls, namely tsik, egg, ocks, chatter, and seep. At 6 weeks of age, mature contact calls predominated the control animal’s vocalization. However, in ACC-lesioned animals, this aspect of social vocal behavior was substantially reduced. This difference emerged gradually after the surgery and was only evident at 6 weeks of age (Figure 3E; χ²(2)=12.73, p=0.026). By the sixth week, the social vocal repertoire of the lesioned animals was altogether different from the control animals, with a much smaller proportion of social contact calls. We examined cries separately as immature social contact calls since parents are generally more responsive to infant cries, and this socially reinforces vocal development (Gultekin and Hage, 2017; Gultekin and Hage, 2018). We found, however, that infant cry rates reduced substantially over the course of 6 weeks. In fact, the controls stopped crying at postnatal week 5 (percent cries compared to all call types: controls, week 5~3%, week 6~0%; ACC, week 5~20%, week 6~8.4%). This might be related to how cry-calls transition from immature to adult-like calls during babbling, thought to be accelerated with parental feedback (Gultekin and Hage, 2018).

To further understand the effect of the ACC lesion on the normal distribution of calls, we investigated the call transition probabilities between contact calls, cries, and other calls (Figure 3F). In contrast to the control animals, whose repertoire was dominated by social contact calls, the ACC lesion group showed frequent transitions mostly to other non-contact call types (u-index Wilcoxon test, p=0.055). These data suggest, therefore, that the ACC mediates developmental changes within the first 6 weeks of life that lead to the dominant production of isolation-induced contact calls and the gradual reduction of cries and other calls.

We examined the characteristics of vocal sequences to learn more about how early life ACC lesions might influence the acoustic signals that marmosets potentially relay to distantly located family members or other conspecifics when socially isolated. We focused on phee syllables, which are discrete elements or components of a call separated by very short intervals (Devinsky et al., 1995; Agamaite et al., 2015). Thus, a sequence of phee calls may comprise multiple syllables (Figure 4A). The functional significance of syllables is not clearly understood, but a change in the number of syllables or their acoustic characteristics might feasibly alter the message conveyed to a family that cannot be seen or heard. This is especially important to young infant monkeys that are naturally demanding of attention, and even more so if isolated. The number of syllables per phee call was highly variable among the animals, ranging from 1 to 8 syllables per phee.

The ACC lesion did not greatly affect the phee syllable count. Aside from a transient decline in the number of syllables in the week after the surgery (postnatal week 3: Wilcoxon test p=0.042), these animals showed the normal preferred range of 3–4 phee syllables at later postnatal weeks (Figure 4B). Even in the sixth postnatal week, when the proportion of phee and other contact calls was much lower in the ACC-lesioned animals, the number of syllables in those phee calls that were issued was similar to the control group.

However, other phee call variables were affected by the lesion. For example, the duration of phee syllables was shortened in ACC-lesioned animals (χ²(5)=13.27, p=0.021), particularly in the later syllables of a multisyllabic phee. This effect emerged gradually and was most pronounced when the animals were 5–6 postnatal weeks (Figure 4C). Likewise, the amplitude of phee syllables was also affected by the ACC lesion (χ²(5)=48.178, p<0.0001) with lesioned animals making louder phee calls (Figure 4D). For each phee syllable, the amplitude difference between groups increased until postnatal week 5 and then disappeared at postnatal week 6 (postnatal week 4: β=3017.75, 95% CI [1831.61, 4203.89] p<0.001; postnatal week 5: β=3719.00, 95% CI [2389.88, 5048.12], p<0.001). In line with the increase in amplitude, the peak frequency of each phee syllable was also lowered by the ACC lesion (Figure 4E). This change occurred soon after the lesion (postnatal week 3: β=–488.63, 95% CI [–702.40, –274.85] p<0.001; postnatal week 4: β=–384.61, 95% CI [–607.98, –161.24], t(513) = –3.38, p<0.001).

Finally, we examined the entropy of phee syllables as a measure of vocal complexity (Kershenbaum, 2013). High entropy in multisyllabic phees would indicate that these vocalizations are diverse, variable, and unpredictable. We found that animals with ACC lesions exhibited lower entropy in phee syllables relative to controls as early as postnatal week 3 (χ²(5)=34.528, p<0.0001) (Figure 4E), suggesting that their vocal sequences were simple, less diverse, and stereotyped. Together, our data suggest that the ACC lesion compromised the normal development of the phee signature for each monkey by making them shorter, louder, and monotonic.

All procedures accorded strictly with the Guide for the Care and Use of Laboratory Animals and were approved by the Animal Care and Use Committee of the National Institute of Mental Health (#SBN-02). A total of 10 marmosets (Callithrix jacchus), five males and five females, all born in captivity, were used in this study. Five infants received ACC lesions at 14–16 postnatal days of age. Five others served as age-matched controls. The infants were raised by parents and siblings in family groups comprising four to six members and housed in temperature-controlled rooms (~27°C), 50–60% relative humidity under diurnal conditions (12 hr light:12 hr dark). Food and water were available ad libitum, supplemented with fresh fruit or vegetables. One animal showed sparing of the ACC lesion in one hemisphere. The final sample size for the behavioral data was n=5/per group. For the MRI and histological data, the final sample size was n=4/per group.

We first obtained a reference MRI scan using a 14-day-old ex vivo sample (Figure 1A). A T2-weighted scan was obtained using 7T Bruker Biospin MRI platform with an eight-channel volume coil. Using ParaVision Acquisition 6.0.1, the following echo sequence was used to acquire a three-dimensional volume of the infant marmoset brain: TR = 400, TE = 72 ms, flip angle = 90 degrees, matrix size = 256 × 256 × 214, resolution = 0.15 mm isotropic, number of averages = 8, number of repetitions = 1, and the total scan time was 3 hr. The scan was aligned horizontally by rotating the image until the anterior and posterior commissures were positioned at the same height and water-filled ear bars were used to obtain the interaural reference. We then used ITK-SNAP (Yushkevich et al., 2006) to identify the ACC at the coronal planes before the genu of the corpus callosum to the level of the anterior commissure. The coordinates were calculated relative to the ear bars and midline references, both of which were visible on the scan. The resulting 14-day-old marmoset scan served as a template atlas to calculate injection coordinates to target the rostral portion of the dorsal ACC (24a and 24b), bilaterally, in all marmosets. We calculated five injection coordinates for each hemisphere: (1) AP: 10.7 mm, ML: ±0.7 mm, DV: –2.9 mm; (2) AP: 10.7 mm, ML: ±0.7 mm, DV: –4.5 mm; (3) AP: 9.5 mm, ML: ±0.7 mm, DV: –3.5 mm; (4) AP: 8.5 mm, ML: ±0.7 mm, DV: –3.6 mm; (5) AP: 7.5 mm, ML: ±0.7 mm, DV: –3.5 mm.

The entire surgical procedure was performed under aseptic conditions in infant monkeys that were 14–16 days of age. During surgery, monkeys received isotonic fluids. Heart and respiration rates, body temperature, blood pressure, and expired CO₂ were monitored throughout the procedure. Pre- and postoperatively, monkeys received nonsteroidal anti-inflammatory drugs (meloxicam, 2 mg/ml, s.c.) to reduce swelling. The monkey was first immobilized with an anesthetic dose of alfaxalone (10 mg/kg, i.m.) combined with diazepam (5 mg/kg, i.m.). In this state, the infant’s head was shaved, and vital electrodes were secured on the infant’s chest. Temperature was measured with a rectal probe. The infant’s head was then secured in a small animal stereotaxic frame (Stoelting Company, IL, USA) attached to a custom-built stage fitted with eye bars, ear bars, and a pallet bar to accommodate the small head. Once the head was secured in the frame, anesthesia with isoflurane gas (1–2% to effect) was provided through the custom-fitted mask (Figure 1B). An integral part of the pallet bar was a gas hole (0.6 mm diameter) that ran along the length of the bar and connected, via tubing, to a vital monitor to measure small end-tidal CO₂ volumes.

Following a midsagittal incision, the scalp was retracted, and a craniotomy was made above the target coordinates of the brain. A 5 µl syringe (33 gauge, Neurosyringe, Hamilton Company, Reno, NV, USA) was used to administer bilateral injections of 0.12 M NMDA (M3262, Sigma-Aldrich) dissolved in sterile-filtered saline into the ACC (0.5 µl per injection site). Each injection was made over 2 min and the injector remained in place for an additional 4 min for dispersion. When all injections were complete, the scalp was closed with intradermal absorbable sutures, and the infant was allowed to recover in an intensive care unit that was void of extraneous sensory stimulation (e.g. excessive bright lights and loud noise). During recovery, marmosets received a combination of Esbalic and Enfamil (3:1 ratio) infant formula every 2–3 hr. When fully awake, each infant was returned to its family unit. A total of five marmosets received the neonatal cingulate lesion. Another five marmosets served as controls: two received saline injections (shams), one received a craniotomy only, and another two were unoperated.

Each infant was placed in a temperature-controlled incubator set to ~38°C (ThermoCare, CA, USA), and the emitted vocalization was recorded for 5 min. Sound recordings were acquired using a cardioid microphone (Sennheiser ME 64, Sennheiser, Wedemark, Hanover, Germany) that was placed on the side of the incubator (Figure 1C). The microphone was connected to a computer, and recorded sounds were digitized at a sampling frequency of 44 kHz using Raven Lite software (Cornell Lab of Ornithology, Ithaca, NY, USA). Due to a variety of extraneous factors beyond our control, including restrictions due to the COVID-19 pandemic, the exact day and number of recordings differed between monkeys. Therefore, recording sessions from each infant were grouped by week. All recording sessions were conducted without the presence of investigators in the recording room. The infant was then returned to its family unit.

The spectrogram of each audio file was obtained and visually inspected using Raven Pro 1.6 (Cornell Lab of Ornithology, Ithaca, NY, USA). Spectrograms were generated with a Hann window of 512-sample points to filter the signal at 3 dB bandwidth of 124 Hz (e.g. Figure 2A and B). The calls were manually classified by a defined classification system (Bezerra and Souto, 2008; Pistorio et al., 2006). To identify call types in the spectrogram of a recording, Raven software features, such as amplitude waveform, spectrogram, and audio playbacks, were used. Six major call types (phee, trill, trill-phee, twitter, cry, and complex calls) were identified from spectrograms, along with other minor call types (tsiks, chatter, egg, ock, seep). In some cases, when trill-phees and phees looked similar in spectrograms of a recording, acoustic parameters such as entropy were used to carefully classify calls. Complex calls comprised of vocalizations with elements from at least two different simple call types such as trill-twitter or twitter-phee, etc. From each recording, call types were manually annotated by three trained investigators (inter-rater reliability >80%). The spectrograms were used to obtain acoustic measurements such as peak frequency, RMS amplitude, and aggregated entropy and exported for further analysis.

Call structures such as inter-call intervals, syllable duration, and number of phee syllables were analyzed using a custom-written R script (Nagarajan G, https://github.com/Guru-learn/Cingulotomy_Vocalization; copy archived at Guru-learn, 2025; Nagarajan, 2024). Acoustic analyses were performed only for phees, which served as the major call type because of their abundance during postnatal weeks. In some cases, vocalization quantity and amplitude were largely suppressed for several minutes after handling by the experimenters. Consequently, analysis of each recording began after 2 min had elapsed.

Gross ACC volume was measured from anesthetized MR scans performed at approximately 8 months of age. Anesthesia was induced with 5% isoflurane. The animals were then placed in an MR-compatible cradle where their heads were secured using ear bars. Isoflurane was maintained at 1.5–2.5%, and vitals were monitored with V9004 Series Capnograph Monitor (San Clemente, CA, USA). We obtained T2-weighted scan (n=4 for control and n=4 for ACC) using the MR procedure described above with the following echo sequence: TR = 30, TE = 48, matrix size = 144 × 144 × 128, resolution = 0.25 mm isotropic, number of averages = 8, number of repetitions = 1. The rostro-caudal extent of the ACC was segmented and measured using ITK-SNAP 4.0. A representative sagittal view of the lesioned and non-lesioned ACC can be seen in Figure 1E. Voxels containing ACC were carefully labeled from the anterior to posterior slice of the MR scan for each subject.

At approximately 24 months of age, the marmosets were euthanized and perfused with 0.1 M PBS followed by 4% paraformaldehyde. The brains were extracted and cryoprotected in 0.1 M phosphate-buffered sucrose (in steps of 10%, 20%, and 30% wt/vol). The brains were partitioned along the midline, and right hemispheres were used for further histological processing after sectioning in coronal orientation on a sliding microtome and cryostat into 40 μm sections.

The immunohistochemistry (IHC) was performed on two series of free-floating sections. Initially, from each of the eight brains, six sections each were collected around three rostrocaudal planes at the following approximate locations (in reference to the interaural axis): +13.30 mm (target brain region: the ACC); +9.20 mm (target brain region: the AMY); and +2.05 mm (target brain region: the PAG matter). Subsequently, these 18 sections were divided into two IHC series of nine sections each, where three sections covered each target brain area for separate processing. The first IHC series was processed to visualize major cell classes that are likely to be affected by a brain lesion (neurons: primary antibody against NeuN, astrocytes: primary antibody against the glial fibrillary acidic protein [GFAP], microglia/macrophages: primary antibody against the ionized calcium-binding adaptor molecule 1 [Iba1]). The second IHC series was processed to visualize and assess the ratio of GABAergic neurons to all neurons: such changes in the relative number of inhibitory neurons could indicate local downstream neuroplasticity in reaction to the ACC lesion, as GABAergic neurons are primarily responsible for local inhibition. In the second IHC series, we visualized the distribution of NeuN, as well as of neurotransmitter GABA and GAD67, a rate-limiting enzyme in GABA synthesis that produces more than 90% of GABA in the central nervous system.

The first IHC series was incubated in a cocktail of the following primary antibodies for 60 hr at 4°C: NeuN (chicken, ABN91, Millipore Sigma), 1:500 dilution; GFAP (goat, SAB2500462, Millipore Sigma), 1:1500 dilution; Iba1 (rabbit, 019-19741, Fujifilm Wako Chemicals), 1:1500 dilution. Then, after several washes, the sections were incubated in a cocktail of the following secondary antibodies for 2 hr at RT: 1:200 donkey anti-chicken Alexa Fluor 488 (703-545-155, Jackson ImmunoResearch), 1:500 donkey anti-goat Alexa Fluor 594 (A21207, Invitrogen), 1:400 donkey anti-rabbit Alexa Fluor 680 (711-625-152, Jackson ImmunoResearch). Additional antibodies used in the second IHC series were as follows. Primary: GABA (rabbit, A2052, Millipore Sigma), 1:600 dilution; GAD67 (mouse, MAB5406, Millipore Sigma), 1:500 dilution. Secondary: 1:500 donkey anti-rabbit Alexa Fluor 594 (A21207, Invitrogen), 1:400 donkey anti-mouse Alexa Fluor 680 (A10038, Invitrogen). Finally, the sections were mounted on a glass slide, air-dried, and coverslipped with DEPEX mounting media (13515, Electron Microscopy Sciences, Hatfield, PA, USA).

Myelinated fibers were stained with aurohalophosphate-based Black-Gold II (Histo-Chem Inc, Jefferson, AR, USA) as shown in Schmued et al., 2008, according to the manufacturer’s protocol and coverslipped with Permount (SP15, Fisher Scientific).

Fluoro-Jade C (Histo-Chem Inc) stain was used to visualize degenerating neurons according to the manufacturer’s protocol in the target brain regions of ACC, AMY, and PAG. There was no observable signal of neurodegeneration in the studied regions.

Each histological section was digitized at 0.65 μm resolution (×10 magnification) using a Zeiss Axioscan microscope slide scanner. Images were then split into a separate channel for each fluorophore. Cell detection and counting were done with an open-source QuPath software 0.3.2 (Bankhead et al., 2017). As each fluorescence channel was analyzed separately, the loci of immunofluorescence that were counted do not necessarily correspond to unique cells, especially for microglia and GABA channels where the fluorescent signal was more diffuse in appearance. Cell segmentation in brain regions downstream to the lesion site was done using the random trees algorithm embedded into QuPath. Cell segmentation of the lesion site was done by custom-trained Cellpose models (doi: 10.1038/s41592-022-01663-4). Although the researcher performing cell counts was blinded to the marmosets’ identity, it was possible to identify the site of the lesion and determine the animals’ group membership. Raw cell counts were transformed into cell densities per mm2 to account for size differences in region of interest (ROI) areas.

All analyses were performed using R 4.2.3 (https://www.R-project.org/). For vocalization analysis, recording sessions were averaged by week, and unless otherwise noted, the data are represented as the mean values with mean ± confidence interval. Recording sessions obtained before the surgery were grouped as postnatal week 2 (pre-surgery), and all the recordings after the surgery were binned into postnatal weeks 3–6 (post-surgery). Due to COVID-19-related restrictions, recordings for some infants could not be extended beyond postnatal week 3.

Packages in the tidyverse library (Wickham et al., 2019) were used for data processing and analyses. Linear mixed effect models (LMMs) were used to analyze postnatal datasets, and these models were fitted with the lmer() function in lme4 package (Bates et al., 2015). For the estimation of coefficients, the maximum likelihood method was used. Models were fitted with postnatal weeks and experimental groups as fixed factors, including an interaction term between treatment and postnatal weeks to assess whether the effect of weeks differed across treatment groups. To account for inter-individual variability, each monkey was modeled as a random effect. For multisyllabic phee analysis, syllables were nested within monkeys’ random effect. Models with and without lesions were used to test the effect of the ACC lesion, and the lesion effect was considered significant at an α of 0.05 (Winter, 2013; Brown, 2021). Model assumptions were tested using the check_model() function available in the performance package. Log transformation was performed on some dataset to meet LMM assumptions. When the normality assumption was violated, a non-parametric test (Wilcoxon test) was also used. Graphs were created using the package ggplot2.

For analysis involving immunofluorescence, there were some inhomogeneities in the spatial distribution of IHC signal. The raw immunopositive detections for specific markers were normalized by NeuN cell counts that were obtained in the same QuPath processing pipeline. The final measure for each antigen-specific immunopositivity count was the ratio computed from the number of antigen-positive detections divided by the sum of the antigen-positive detections and NeuN-positive cells. This transformation bounded the possible antigen-specific detections between 0 and 1 and allowed for parametric modeling with beta distribution. Cell counts were fitted using the glmmTMB software package in R statistical computing environment (https://github.com/kissmyjazz/cingulotomy-histology, copy archived at Matrov, 2025). As there were three sections per animal for each ROI, these were modeled as random effects. Widths of major fiber tracts were modeled by a linear regression (function lm in base R). Final p-values were adjusted for multiple comparisons with the false discovery rate method for each antigen.