Arousal dynamics drive vocal production in marmoset monkeys

12 Pages·2016·1.83 MB·English

JNeurophysiol116:753–764,2016.
FirstpublishedJune1,2016;doi:10.1152/jn.00136.2016.
Arousal dynamics drive vocal production in marmoset monkeys
Jeremy I. Borjon, Daniel Y. Takahashi,* Diego C. Cervantes, and Asif A. Ghazanfar*
PrincetonNeuroscienceInstituteandDepartmentofPsychology,PrincetonUniversity,PrincetonNewJersey
Submitted15February2016;acceptedinfinalform31May2016
Borjon JI, Takahashi DY, Cervantes DC, Ghazanfar AA. bonds(Kulahcietal.2015),positioninasocialhierarchy(Kitchen
Arousal dynamics drive vocal production in marmoset monkeys. J etal.2005),distancefromconspecifics(Choietal.2015;Fischer
Neurophysiol 116: 753–764, 2016. First published June 1, 2016;
et al. 2001), seeing a predator (Seyfarth et al. 1980), hearing
doi:10.1152/jn.00136.2016.—Vocal production is the result of inter-
conspecific calls in general (Ghazanfar et al. 2001; Takahashi et
acting cognitive and autonomic processes. Despite claims that
al.2013),andhearingcallsfromaspecificindividual(Millerand
changesinoneinteroceptivestate(arousal)governprimatevocaliza-
tions,weknowverylittleabouthowitinfluencestheirlikelihoodand Thomas2012).Thesefindingsarebolsteredbyrecentneurophys-
D
timing. In this study we investigated the role of arousal during iologicalstudiesimplicatingneocorticalstructuresintheproduc- o
naturallyoccurringvocalproductioninmarmosetmonkeys.Through- tionofvocalizations(Coudéetal.2011;Hageetal.2013;Hage w
n
outeachsession,naturallyoccurringcontactcallsareproducedmore andNieder2013).Whereastheaboveexternalfactorshavebeen lo
a
quickly,andwithgreaterprobability,duringhigherlevelsofarousal, exhaustively studied, we know very little about how internal, de
asmeasuredbyheartrate.Onaverage,weobservedasteadyincrease d
ipnrohdeuacrttiorna,tethe2r3eissbaesfhoarerptahnedpsrtoedepucctaiordniaocfdaeccealellr.atFioonllolawsitninggc(cid:2)al8l autOonnoemaiuctpornoocmesiscesnmeravyouinsflsuyesntceemvofcuanlcptiroonduicstiotonicnonptrrimolattehse. from
s.Thedynamicsofcardiacfluctuationsaroundavocalizationcannot state of arousal. Arousal is relevant for a range of behaviors h
be completely predicted by the animal’s respiration or movement. (Pfaff 2006), and an animal would be said to exhibit a high ttp
Moreover,thetimingofvocalproductionwastightlycorrelatedtothe arousal state if it is more alert to sensory stimuli, more ://jn
phase of a 0.1-Hz autonomic nervous system rhythm known as the motoricallyactive,andmorereactive(Gareyetal.2003;Pfaff .p
h
Mayer wave. Finally, a compilation of the state space of arousal y
2006).Inthedomainofvocalbehavior,thefocushasbeenon s
dynamics during vocalization illustrated that perturbations to the io
resting state space increase the likelihood of a call occurring. To- linking presumptive arousal level changes to changes in be- lo
gether, these data suggest that arousal dynamics are critical for havioral context and, subsequently, to changes in the acoustic gy
spontaneousprimatevocalproduction,notonlyasarobustpredictor properties of vocalizations. For example, in tree shrews, de- .org
of the likelihood of vocal onset but also as scaffolding on which creasing physical distance between a male and female is b/
behaviorcanunfold. thought to increase arousal levels in the female, resulting in a y
1
higher rate of vocal output (with higher fundamental frequen- 0
anterior cingulate; autonomic nervous system; default mode; intero- .2
ception;Mayerwave cies; Schehka et al. 2007). Along the same lines, levels of 20
“response urgency” can be modulated by distance from a .3
predator,andthisislinkedtochangesintheacousticstructure 3.5
of alarm calls in squirrels (Owings and Virginia 1978), mar- o
n
NEW & NOTEWORTHY mots (Blumstein and Armitage 1997), and meerkats (Manser A
2001). Acoustic changes in cat (Scheumann et al. 2012), bat ug
We investigated the role of arousal during naturally oc- (Bastian and Schmidt 2008), and horse (Briefer et al. 2015) us
curringvocalproductioninmarmosetmonkeys.Thetiming vocalizations were also reported as a response to context- t 1
ofvocalproductionwastightlycorrelatedtoanautonomic 7
nervous system rhythm known as the Mayer wave. A state related arousal fluctuations. In primates, changes in arousal , 2
levels are also repeatedly linked to changes in vocal acoustics 0
space of arousal dynamics during vocalization illustrated 1
(Hammerschmidt and Fischer 2008; Owren et al. 2011; Ren- 6
that perturbations to the resting state increase the likeli-
hoodofacalloccurring.Arousaldynamicsarecriticalfor dallandOwren2009).Inbaboons,forinstance,Rendall(2003)
spontaneousprimatevocalproduction,actingasthefoun- examined the grunt vocalizations of females in two different
dation on which vocal behavior can unfold. contexts and, within these two contexts, categorized low and
high arousal states based on patterns of movement. He found
VOCAL PRODUCTION is the result of interacting cognitive and thatvariationaccordingtoarousallevelswasmainlyassociated
autonomic processes, similar to those underlying the produc-
withchangesintherateofcallsproducedandtheirduration,as
tion of facial expressions or other communicative signals
well as changes in the fundamental frequency (Rendall 2003).
(Gothard 2014). In nonhuman primates (hereafter, primates),
In the current study, we address a different question with
we know that many external factors influence the timing and
regard to arousal and vocal production: What mechanism
likelihood to produce a vocalization (Seyfarth and Cheney
drives an individual primate to produce a vocalization at all?
2003). Such factors include an individual’s recent history of
To investigate this, we needed a model system that readily
social interactions (Cheney and Seyfarth 1997), strength of
produces vocalizations without the necessity of an external
event such as a conspecific vocalization, predator, or other
*D.Y.TakahashiandA.A.Ghazanfarareco-seniorauthorsofthisarticle. overt sensory trigger. Such a context would allow us to focus
Addressforreprintrequestsandothercorrespondence:Addressforreprint
solelyonhowchangesinarousalmayormaynotbelinkedto
requestsandothercorrespondence:A.A.Ghazanfar,PrincetonNeuroscience
vocal production. The calling behavior of marmoset monkeys
Institute and Dept. of Psychology, Princeton Univ., Princeton, NJ 08544
(e-mail:[email protected]). (Callithrixjacchus)meetsthiscriterion(BorjonandGhazanfar
www.jn.org 0022-3077/16Copyright©2016theAmericanPhysiologicalSociety 753
754 MAYER WAVE AND VOCAL PRODUCTION
2014). When alone, marmoset monkeys vocalize readily and separate calls, whereas gaps shorter than or equal to 1 s indicated
spontaneously (Bezerra and Souto 2008; Takahashi et al. syllables from the same call. After this procedure, we manually
2013). Typically, if a conspecific is within hearing range, the verified for each call whether the automatic routine correctly identi-
two individuals will take turns exchanging calls (Takahashi et fied single phee calls or combined multiple calls, using the 1-s
separationcriteria.Furthermore,wecountedthenumberofsyllables
al. 2013), but if not, the isolated marmoset will continue
of each call, and if there was any mismatch between the automatic
calling. Moreover, in this undirected context, nearly all the
onset-offset detection and the call observed in a spectrogram, we
callstheyproduceareasingletypeofcontactcall,the“phee”
markedtheonsetandoffsetofthecallmanually.Atotalof427calls
call.Itisahigh-pitchedvocalizationwithoneormorelargely
were produced across a span of 26 sessions (for monkeys 1–4: 192,
identical“syllables”(BezerraandSouto2008;Takahashietal. 29,120,and86calls,respectively).
2013).Becausecontextcontrolsthecalltype,thisallowsusthe Heartrateanalyses.Surfaceelectrodesalongboththeventraland
clearest possible means to investigate how arousal may influ- dorsal thorax of the marmoset can each pick up both heart rate and
ence vocal production. respiratorysignals.Thestrengthofthesesignalscanvarythroughout
asessionasthepositioningofthesurfaceelectrodeschangeswiththe
animal’smovement.Therefore,wesoughttoscreenthequalityofthe
MATERIALS AND METHODS data and determine which channel exhibited the cardiac signal with
the largest signal-to-noise ratio on a session-by-session basis. This
D
Four adult common marmosets (Callithrix jacchus; 3 males and 1 channelwasusedforallheartrateanalysesforthatsession.Because o
female,averageage5.1yr)housedatPrincetonUniversitywereusedin automated methods of detecting and removing noise from these w
n
thisstudy.Themediannumberofsessionstheyparticipatedinwasseven selected channels were suboptimal, we manually identified and iso- lo
a
sessionsperanimal(monkeys1–4:9,3,9,and5sessions,respectively). latedmotionartifactsorsignalcutoffs.Tominimizebias,wedidthe d
e
Allmarmosetslivedwiththeirpair-bondedmatesinfamilygroupsand following: for each session, signals from both EMG channels were d
were born in captivity. They were fed daily with standard commercial dividedinto10-ssegmentsandsignalpairswerepresentedinrandom fro
chowsupplementedwithfreshfruitsandvegetablesandhadadlibitum order for visual inspection. Regions exhibiting signal loss or motion m
access to water. The colony room was maintained at a temperature of artifactswerereplacedwithNaNs(not-a-number;MATLAB). h
(cid:2)27°Cwith50–60%relativehumidityanda12:12-hlight-darkcycle. Followingthisscreening,wedown-sampledthedatafrom40kHz ttp
Allexperimentalsessionswereconductedduringdaylighthoursbetween to 1,500 Hz to extract the cardiac signals. These signals were then ://jn
1200and1800.Allexperimentswereperformedincompliancewiththe high-pass filtered at 15 Hz to preserve the rapid waveform of the .p
h
guidelines of, and were approved by, the Princeton University Institu- heartbeat.Toremovelinenoise,theresultingsignalwasnotchfiltered y
s
tionalAnimalCareandUseCommittee. at60Hz.Heartbeatsweredetectedusinganadaptivethresholdof1-s io
Electromyography.Fortheexperiment,theanimalwasplacedina durationtofindcardiacspikesgreaterthanthe95thpercentileofthe lo
g
testing box. The testing box was made of Plexiglas and wire in a amplitudeateachsecondofthesignal.Occasionally,thedetectionof y
triangular prism shape (0.30 (cid:3) 0.30 (cid:3) 0.35 m). To record the heartbeatswouldpickupaspuriousspikeclosetotheactualheartbeat. .org
electromyographic (EMG) signal, we used two pairs of Ag-AgCl This would result in a series of spikes occurring with an interspike b/
surface electrodes (Grass Technology). Tethered electrodes were timingfasterthan100ms,correspondingtoabiologicallyimplausible y
1
sewnintoasoftelasticband,whichwasclaspedaroundtheanimal’s 600beats/min.Wethereforeimplementedaspeedthreshold,whereby 0
thorax.Onepairofelectrodeswasaffixedtothechestareaclosetothe 2spikesoccurringinsuccessionquickerthan100msweresubstituted .2
2
heartandthesecondpairplacedontheback,closetothediaphragm. withasinglespikelocatedatthemidpoint.Itisalsopossibleforthe 0
.3
To improve the signal-to-noise ratio, we applied ECL gel on the EMGrecordingtomissaheartbeat.Ifanyinterbeatintervalbetween 3
surfaceofeachelectrode.Ifneeded,marmosetswereshavedaround spikeswaslargerthan400ms(150beats/min),whichistheminimum .5
thethorax.Eachpairofelectrodeswasdifferentiallyamplified((cid:3)250) heart rate reported in conscious, unrestrained adult marmosets (Sch- on
with the resulting signal sent to a Plexon Omniplex, where it was nellandWood1993),wereplacedtheinterveningsignalwithNaNs A
u
digitizedat40kHzandsenttoapersonalcomputerfordataacquisi- toaccountforthesignalloss.Tocalculateheartrate,weconstructed g
tion.Recordedsessionswerebetween20and50min(means(cid:4)SD, abinaryseriesofheartbeatcountsandconvolvedtheresultingseries us
monkeys 1–4: 21.86 (cid:4) 11.70, 31.76 (cid:4) 5.51, 22.63 (cid:4) 12.32, and witha1-sGaussianwindow.Thismethodisoftenusedinestimating t 1
7
15.44(cid:4)4.86,respectively). the instantaneous rate of action potential occurrences in neurophysi- , 2
Recording vocalizations. Vocalizations were recorded using a ologicaldata(ShimazakiandShinomoto2010). 0
1
Sennheiser MKH416-P48 microphone suspended 0.9 m above the Respiratoryrateanalyses.Thesamenoisecontrolprocedureused 6
testingbox.ThemicrophonesignalwassenttoaMackie402-VLZ3 to determine the best channel of cardiac activity was used to deter-
linemixerwhoseoutputwasthenrelayedtothePlexonOmniplexfor mine which channel demonstrated the best respiration signal. Seg-
acquisition. This ensured the timing accuracy of the vocal signals ments of signals that did not demonstrate a respiratory signal were
relativetotheEMGsignals.Althoughmarmosetsproduceanumber excludedfromfurtheranalysisbybeingreplacedwithNaNs.Sincewe
of distinct vocalizations in a number of different contexts (Bezerra wereinterestedinphysiologicalsignalsrelatedtorespiration((cid:5)3Hz),
andSouto2008),99%ofthevocalizationsrecordedintheundirected wedown-sampledthedatafrom40kHzto50Hz.ToremovetheDC
context of social isolation used in this study were phee calls, the offset, we high-pass filtered the signal at 1 Hz. This resulting signal
species-typicalcontactcall.Weusedthesamecriterionweestablished was further low-pass filtered at 4 Hz to isolate respiratory activity.
in previous work for computationally defining and segmenting phee Because mammals, including primates, produce vocalizations using
calls and their syllables (Takahashi et al. 2013). This and all subse- expiration-drivenvocalfoldmovements(FitchandHauser1995),we
quentanalyseswereconductedinMATLAB2015a.Thesegmentation wantedtomakesurethattherespiratorysignalalwayshadaconsis-
routine automatically detected the onset and offset of any acoustic tentphaserelationshipwithvocaloutput.Thatis,wewantedtokeep
signalthatdifferedfromthebackgroundnoiseataspecificfrequency the expiratory phase of the respiration signal in a consistent orienta-
range. To detect the differences, we bandpassed the entire recorded tion relative to vocal production (so that the beginning of expiration
signal between 5 and 8 kHz for each session. This frequency band wasthepeakoftherespiratorycurve).Todothis,weensuredthatthe
capturesthefundamentalfrequencyofpheecalls.Wethencompared slope between the point of respiration at call onset and the point of
the amplitude of the signal at this frequency band for the periods respirationatthepeakcallamplitudewasnegative.Thephasecoor-
without calls and during a call. A simple threshold was enough to dinatesoftherespiratorysignalwerethendeterminedbycalculating
distinguish both periods. Onset-offset gaps longer than 1 s indicated theangleoftheHilberttransformoftheresultingsignal.Tocalculate
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
MAYER WAVE AND VOCAL PRODUCTION 755
respiratoryrate,wefirstdetectedthepointsintimewheretheangleof 1,000 times. The 95% threshold for significance corresponds to the
theHilberttransformwas0,representingthebeginningofexpiration. 2.5thand97.5thpercentilesofthebootstrappedmedians.
These time points were converted into a binary representation and Confidence intervals for the medians of heart and respiratory rate
convolved with a 2-s Gaussian window. Again, this method is often and motor activity were generated by randomly resampling, with
used in estimating the instantaneous rate of spike occurrences replacement, from the signals used to calculate the plotted medians.
(ShimazakiandShinomoto2010). Wecalculatedthemedianoftheresampledsignalsandfitittoacubic
Calltimingandprobabilitymeasures.Wewantedtounderstandthe spline using csaps in MATLAB. We repeated this process 1,000
relationshipbetweenheartrateandcalltimingonasessionlevel.That times. The 95% confidence interval corresponds to the 2.5th and
is, we wanted to determine how quickly a call will occur given the 97.5thpercentilesofthebootstrappedmedians.
currentheartrateatanypointduringasession.Toobtaintheheartrate Linearpredictionofheartrateusingrespirationrateandthemotor
ataresolutionof1s,wedividedthedatainto1-sbinsandcomputed index.Wesoughttolinearlypredictheartratefromongoingrespira-
theaverageheartrateforeachbin.Thereareconsiderabledifferences tory activity. We first calculated the cross-correlation of heart and
inheartratebetweenindividuals(Cacioppoetal.1994).Tonormalize respiratory rate using the median of the data set, and then took the
changes in the magnitude of heart rate between our marmoset sub- maximalandminimalpeakafter0sdelay.Thesepointsindicatethe
jects, we converted the 1-s averages into percentiles (from the 0th times at which respiratory rate maximally correlates to heart rate.
percentile up to the 100th percentile, in bins of 1 percentile). Call Alignedtocallonset,thefirstpositivepeakoccursat5s,whereasthe
timingrelativetoeachbinofdatawasdeterminedbymeasuringthe negative peak occurs at 23.34 s. Aligned to call offset, the first
D
intervalfromeachbintotheonsetoftheclosestcall.Tovisualizethe positive peak occurs at 5.78 s, whereas the negative peak occurs at o
averageprobabilitycurve,wefitthedatatoacubicsplineusingcsaps 21.48s.Webuiltlinearpredictorsofheartratefromrespiratoryrate w
n
inMATLAB. starting from these points in time. To build them, we calculated the lo
a
Wealsowantedtoknowtheprobabilityofobservingacallgiven linearregressionbetweenheartrateandrespiratoryrate.Wecreated d
e
a certain heart rate from any point within the session. Using the 1-s anintervalofpredictedheartratebyusingabootstrapprocedurefor d
averages, we measured the probability of observing a single call the data aligned to call onset and call offset. For this, we uniformly fro
within23sofeachbin.Thistimepointcorrespondstotheminimum andrandomlysampledsegmentsofheartrateandthecorresponding m
average time interval from each heart rate percentile to the closest respiratory rates. For each pair of segments, we then calculated the h
subsequent call, as described in the preceding paragraph. To better predictedheartratecurveusingtheassociatedrespiratoryratecurve. ttp
visualize the probability of observing a single call within 23 s, we This resampling procedure was repeated 1,000 times. The 95% ://jn
grouped the data at 5 percentile steps. To show the slope of the confidenceintervalcorrespondstothe2.5thand97.5thpercentilesof .p
h
correlation, we used linear regression to plot the line of best fit. We thepredictedheartrate. y
s
calculatedthecorrelationusingcorrinMATLABandranaSpearman Weconstructedananalogouslinearpredictorusingtheaverageof io
correlation between the averaged data points and heart rate the motion index and the median of the heart rate data. Because the lo
g
percentiles. motionindexisgatheredatasamplingrateof1Hz,wedown-sampled y
.o
Index of motor activity. To determine whether a marmoset was the50-Hzheartratemedianto1Hztomatchthemotionindex.We rg
moving and active during the session, we analyzed the videos re- cross-correlated the down-sampled heart rate median to the mean of b/
cordedduringtheexperiment(23of26sessionswerevideorecorded themotionindex.Alignedtocallonset,thefirstpositivepeakoccurs y
1
at30frames/susingPlexonCineplex).Eachrecordingwassplitinto at6s,whereasthenegativepeakoccursat29s.Alignedtocalloffset, 0
segments of 30 frames. For each segment, we took the absolute thefirstpositivepeakoccursat8s,whereasthenegativepeakoccurs .2
2
differencebetweenthe1stand30thframes.Thisvaluecorresponded at 32 s. We calculated the linear predictor sequence as described 0
.3
tothedifferenceinpixelluminancebetweenthetwoframes.Ahigher above. The 2.5th and 97.5th percentiles of the predicted heart rate 3
value indicates more of a difference, signifying movement. We were up-sampled to 50 Hz from 1 Hz. Down-sampling and up- .5
o
calculated,foreachsession,the90thpercentileofthesepixelvalues samplingwereconductedusingtheresamplefunctioninMATLAB. n
and used that as a cutoff to construct a binary signal. Values higher Probability density and power spectrum density estimates. To A
u
thanthecutoffwereinstancesofmovement,whereasvaluesbelowthe determine the oscillatory nature of marmoset vocal production, we g
u
cutoffwerenot. firstcalculatedthetiminginterval(inseconds)betweenspontaneous s
Wemanuallyscoredthetimingofeachvocalizationbywatchingthe vocalizationsforeverymarmoset.Theintervalwasfromtheonsetof t 1
7
mutedvideo.Weonlymarkedinstanceswherewecouldclearlysee,via one call until the beginning of the subsequent call. Using this input , 2
mouthmovements,thebeginningandendofavocalization.Thebegin- calculated across all animals, we constructed a probability density 0
1
ning of a vocalization was marked at the frame prior to the animal functionusingksdensityinMATLAB. 6
openingitsmouth.Theendofthevocalizationwasmarkedattheframe To visualize the range of frequencies underlying the dynamics of
afterclosureofthemouth.Fromthis,weaccountedfor239vocalizations theheartandrespiratoryratesignal,wecalculatedthepowerspectrum
ofthe427usedinheartrateandrespiratoryrateanalysisin23videos. densityofthesignal25-sbeforeeachcallonsetforeveryanimaland
Becausethevideoframerateisat30Hz,ouridentificationofcalltypes session.Eachsignalwasdetrendedandnormalizedbycalculatingits
hadamaximumresolutionof1/30s.Usingthevideo-markedvocaliza- Zscore.WeusedpburginMATLABtocalculatethepowerspectrum
tiononsets,wewereabletocompiledataforFig.4B. density estimates. To calculate the population power spectrum, we
Populationmedians,individualmedians,andsmoothing.Tovisu- computedtheaverageacrossallcalls.
alize the dynamics of heart rate, respiratory rate, and motor activity To calculate the call rate, we constructed a binary equivalent in
surroundingacallevent,wetookthemedianof40sbeforeandafter length to each of the 26 sessions. This binary was convolved with a
bothcallonsetandoffset,andwefitacubicsplinetothedatausing 1-sbandwidthGaussianwindowandthendetrendedandZ-scored.We
csaps in MATLAB (smoothing parameter 0.3 for the population calculated the average power spectrum density for the resulting
medianand0.01fortheindividualmedians). signals using pburg. To allow comparison between the signals, we
Bootstrappedpermutationtestsandconfidenceintervals.Todeter- normalizedallpowerspectratotheirfirstobservedpeak.
mine whether the changes in heart and respiratory rate and motor Phase calculations. To calculate the phase of the heart and
activity were significant, we performed a bootstrapped significance respiration rate, we calculated the angle of the Hilbert transform
test. For each session, we chose random segments equivalent in for each call. We then calculated a circular phase density plot for
numberandlengthtothecallsproducedinthatsessionfromourdata the phase at call onset using circ_ksdensity in MATLAB. A
set.Themedianofthisrandomlyselecteddatawascalculatedandfit significance test was constructed by first randomly sampling, for
toacubicsplineusingcsapsinMATLAB.Werepeatedthisprocess eachsession,anumberofpointsequalinnumbertothenumberof
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
756 MAYER WAVE AND VOCAL PRODUCTION
calls. We then calculated the phases of these points for every across the entire data set reveals two distinct peaks: a slow
sessionandcalculatedthedistributionofphasesusingthecircular frequency associated with respiration and a higher frequency
density plot as before. We repeated this procedure 1,000 times to
linkedtocardiacactivity(Fig.1B).Anadaptivethresholdwas
obtain the bootstrap confidence interval.
used to identify cardiac spike trains in the high-frequency
StatespaceplotandmultivariateGaussiandistributionfitting.To
band,andthiswasthenusedtomeasureheartratechanges.In
understand the cardiorespiratory dynamics underlying vocal pro-
duction,weconstructedthestatespacefromthemedianheartrate the lower frequency band, respiratory peaks were identified
and respiratory rate during a vocalization. We plotted heart rate using a Hilbert transform and then converted into respiratory
againstrespiratoryratefor25sbeforecallonsetuntil20safterthe rate. As in humans (Cacioppo et al. 1994), marmosets exhib-
call began. ited considerable individual differences in the range of heart
To visualize the scope of cardiorespiratory variability within our
andrespirationrates.Toputthesevaluesinauniformscale,we
data set, we fit the permuted data set used for the bootstrapped
normalized each subject’s heart and respiratory rate into per-
significancetestofheartandrespiratoryratevariabilitytoaGaussian
centiles. Figure 2A shows the different ranges of heart rate
distribution.Wefirstcalculatedthecardiorespiratorydataset’scova-
riance matrix using cov in MATLAB. We then fit a multivariate (monkeys1–4:321(cid:4)75,409(cid:4)39,388(cid:4)45,and382(cid:4)55
normal Gaussian distribution to the permuted data and calculated beats/min,respectively)andrespiratoryrate[formonkeys1–4:
the contours encompassing the 95% confidence interval for the 85 (cid:4) 24, 85 (cid:4) 24, 77 (cid:4) 19, and 83 (cid:4) 27 respiratory
distribution. cycles/min (rpm), respectively]. D
o
If arousal contributes to spontaneous vocal production, we w
n
would predict an interaction between heart rate, a temporally lo
RESULTS a
precise index of arousal state (Gray et al. 2012; Obrist et al. d
e
Increasesinarousalprecedetheproductionofvocalizations 1970),andthetimingandlikelihoodtovocalize.Atotalof427 d
and influence its probability and timing. We used surface spontaneouspheecallswererecordedduringourEMGexper- fro
m
electromyography (EMG) to measure muscle activity around iments.Figure2Bshowsthenegativecorrelationbetweentime h
the thorax of four unrestrained adult common marmosets as ofvocalizationonsetandheartrateforboththepopulationand ttp
they spontaneously produced contact phee calls in an undi- individual monkeys. The onset of call production was quicker ://jn
rectedcontext(i.e.,socialisolation).Thisallowedustocapture following epochs of high vs. low heart rates (for population, .p
ongoingheartratesimultaneouslywithrespiratoryactivity,the Spearmancorrelation,(cid:6)0.978,P(cid:5)0.001).Theexactonsetof hy
s
latterofwhichisdirectlyrelatedtomammalianvocalproduc- callproductionrelativetoheartrateisalsohighlydependenton io
tion. The undirected context contained no external social sig- theindividualanimal,withsomeanimalsproducingcallsmore lo
g
nals, meaning there were no response phee calls produced by frequently than others. To illustrate this, we have provided y.o
anout-of-sightconspecific,aswouldtypicallybethecaseina data for each animal in our study. While high heart rate rg
contact-calling scenario (Takahashi et al. 2013). Figure 1A modulates the timing of vocal onset, it does not provide b/
y
shows the standard position of electrodes on the body and an informationregardingthelikelihoodofacalloccurringafter 1
0
exampleoftherawEMGsignalrecordedduringaone-syllable a given level of cardiac activity. For example, it could be .2
2
phee call. An average power spectrum of the EMG signal that at a heart percentile of 90, a vocalization is 50% likely 0
.3
3
Fig. 1. Measuring respiration and heart rate A B .5
ffraocmeeelleeccttrroodmesyowgerarephaipcp(lEieMdGto)tshigendaol.rsAa:lsaunrd- werB) -10 on A
ventral thorax of an adult marmoset. Signals EMG Po(d -20 ug
wereprocessedthroughadifferentialamplifier 1 2 3 4 5 6 7 u
(Diff.amp)beforebeingsenttothedataacqui- Frequency (Hz) st 1
sition(DAQ)system(PC,personalcomputer). 7
Anindividualexemplarofavocalization(high- PC ~150 ms , 2
lightedinturquoise)withnonvocalizationperi- 0
1
odsisplottedinblack.Theexemplarcomprises Adaptive threshold to detect 6
thevocalization’sspectrogram,theamplitudeof
the vocalization, and the raw data down-sam- DAQ Diff. amp
pledto1,500Hz(Amp.,amplitude;Freq.,fre- Kernel smooth binary into rate
qp2HTuohzpewe)enaeacckrnyasdrs;,dpaoineannc.cuoetr.t,ashrielegnprdnoreiaernnmlsdesoaiinctulytiaitzntpeeigundstgtriemwuhsnepaaiatsiterr)at.sistroodBarlet:yaemtper(oa6dont;pe–suat8(rln0aaHtt–iiezo1nx)n5g-. Freq.(kHz) 48 Heart rate(bpm)348200 1 2Time3 (s) 4 5
p
emplarheartbeatafterfilteringoftherawdatais 1 2 3 4 5 m
shownintheinset.Usinganadaptivethreshold, a 0.02
wtainleedodfectotehncevtesodilgvtnihnaelg,scpiotinkvweseirtethixncgaeeth1de-inssgpGitkhaeeuss9isn5iat%onapkbeeirnrcnaeernyl-, Amp.(n.u.) -101 Resp. -0.020 Find peaks with hilbert transform
resultinginongoingheartrate(bpm,Beats/min).
Afterisolatingtherespiratorysignaloutputthrough 1 2 3 4 5
filtering, we calculated the angle of the Hilbert e Kernel smooth binary into rate
tccrroaonnsvssfeionrtrgemdpotohifnesttsheerpeerpearskepssierianntttieoodnarbseiisgnpnairraayltoawrnyhdeprceeobanykvsoz.levWreode- EMG(mV) 0.04 esp. rat(rpm)1248000
itwitha2-sGaussiankernel,resultinginongoing 1 2 3 4 5 R 1 2 3 4 5
respirationrate(Resp.rate,respirations/min). Time (s) Time (s)
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
MAYER WAVE AND VOCAL PRODUCTION 757
A
M1 M2
R R
160
m) 400 150 esp m)400 esp
art rate (bp 230000 100 iration rate art rate (bp230000 81200 iration rate
He 100 50 (rpm He100 40 (rpm
) )
0 20 40 60 80 100 0 20 40 60 80 100
M3 M4
160 R R
m) 400 esp m)400 150 esp Do
Heart rate (bp 123000000 4812000 iration rate (rp Heart rate (bp123000000 51000 iration rate (rp Frtiaiotgino.sn2..ArRa:etelca,utiamonnudslahttihipveebpedrtwoedneuseicnttyihoepnalroottfsravtfeoo,rcraeelisazpcaih-- wnloaded from
B 0 20 P4e0rcen6ti0le 80 100 m) 0 20 P4e0rcen6ti0le 80 100 m) aagfrnxroeiimmse,na)blvlaou(gcMeaa)il1na–snoMtdnis4tres)etsppppelilrorocattetttionenrtdgiyleahr.ageBtaaeir:nt(asrrvtiagetprheaetgr(cyelee-afnttixtmiiylsee-, http://jn
heart rate for the population with a cubic .p
cal onset (s) 122505000 rp =< -00..090718 ocal onset (s) 246000M210 rp 6 =< 0 -00 ..08011210012355550000M220 rp 6 =<0 -00 ..040511500 sottcAeiipaanmltelclsii.hetznrrCaieaifgtntr:ihefioodatmtnit,.vvhepieApvdrlrsaotouoagtcarbtmealieagldaebphnoiopatlin,pigomtsuaisyenaiilttnaldi.ctsviaiuvtSeonrihtnddvaceuetoaphisarsrrerlttoeaiacrablrraastvaetteberiroraeisalptnphiegetooyberrpwcestoeeetnwdrndfceteveiafmleoonrene-r--. bhysiology.org/
Time from vo 10500 Time from v 135000000M230 rp 6 =< 0 -00 ..0907151001482000M240 rp 6 =<0 -00 ..060611400 Spearmancorrelations. y 10.220.33.5
0 20 40 60 80 100 o
C Percentile heart rate Percentile heart rate n A
u
0.21 0.35M1 rp =< 00..902061 0.14M2 rp =< 00..805081 gus
Probability to vocalize 0000....11113579 rp =< 00..902001 Probability to vocalize00000.....1212255604 M230 6rp0 = = 00..5014100300000000.......00112232605050M240 60rp = = 00..501210180 t 17, 2016
0.12
0 20 40 60 80 100 20 60 100 20 60 100
Percentile heart rate Percentile heart rate
to occur quickly, and at a heart rate percentile of 25, a The baroreceptor pathway does not wholly predict the link
vocalization is also 50% likely to occur but takes longer to between heart rate changes and vocal production. We next
be elicited. Figure 2C shows a positive correlation between examined the dynamics of heart rate during the moments
heart rate and the probability of the phee call being pro- preceding and following the production of a phee call. Phee
duced within 23 s of a given heart rate percentile (for vocalizations can differ in their durations (2.03 (cid:4) 0.96 s); we
population, Spearman correlation, 0.920, P (cid:5) 0.001). Thus therefore plotted the heart rate data aligned to vocalization
the higher the heart rate, the greater the probability of vocal onsetandoffset,separately(Fig.3A).Redlinesindicateepochs
output.Thesedatashowthatboththetimingandprobability that were significantly different from baseline cardiac activity
of vocal production are correlated with heart rate. (i.e., exceeded a 5% significance level). Marmosets exhibited
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
758 MAYER WAVE AND VOCAL PRODUCTION
A
70 70
e) e)
ntil 60 ntil 60
e e
c c
er er
p p
e ( 50 e ( 50
at at
art r 40 95% CI art r 40
e p<0.05 e
H H
Avg. call duration
30 30
-40 -30 -20 -10 0 0 10 20 30 40
Fig.3.Vocalizationsareprecededbyanincrease Time to call onset (s) Time from call offset (s)
B
inarousal.AandB:smoothedpopulationmedi-
ansofongoingheartrate(A)andrespirationrate
(B) plotted with the 95% bootstrapped confi- 70 70
denceinterval(CI)ofthemedian.Redlineindi- e) e) D
csiagtensificvaanluceesteosuttfsoidreca9rd5i%oreospfirtahteorbyovoatrsitarabpilpiteyd. centil 60 centil 60 ownlo
C: cross-correlation plot of the respiration and er er a
h(pcuarhbeoelseselearalap)irdrrnceittrkelataaora)rntatariidatcooteneonpndnn.rseiesiaTscgtdgrrahniaeulcpeatlctilldovtoosseremtaftnffeoslooo(dierogfnttetredhd(e(habageeertlrdnwaaaap)rciyprtatke)hopl).dripegaiTcuacatnrhekltgeoaoesadtsrfii.spronoetoDonosfsmst:micattrthianeheovlesddelneuigoalc(w9topnrns5oieusn%oosrergesff-t Resp. rate (p 345000-40 p9A<5v%0g-3..0 0Cc5aI ll dur-a2t0ion -10 0 Resp. rate (p 453000 0 10 20 30 40 http://jn.pded from
bootstrapped CI of the respiration-predicted Time to call onset (s) Time from call offset (s) h
y
heart rate at its positive (purple) and negative C D s
(green)peak(Pred.,predicted). iolo
1 70 70 g
y
on coefficient 0.05 e (percentile) 5600 e (percentile) 5600 by 10.22.org/
Correlati -0.5 AAlliiggnneedd ttoo oofnfsseett Heart rat 40 +- pprreedd.. ssppaaccee Heart rat 40 +- pprreedd.. ssppaaccee 0.33.5 on
-1 30 30 A
0 10 20 30 40 -40 -30 -20 -10 0 0 10 20 30 40 u
g
Correlation offset (s) Time to call onset (s) Time from call offset (s) u
s
t 1
7
an increase in heart rate before vocal production, followed by arrhythmia (RSA). RSA is a baroreceptor-driven activity aris- , 2
a rapid heart rate deceleration. The heart rate increase began ing from the interaction between the lungs, the nucleus of the 0
1
slowly,startingroughly23sbeforetheonsetofavocalization. solitarytract(NTS),andtheheart.Asignalistransmittedfrom 6
At vocal offset, the steep deceleration in cardiac activity falls the lungs via the vagus nerve to the NTS, where a reciprocal
below the baseline for (cid:2)10 s. In Fig. 3A, the continuing signal is sent back from the nucleus ambiguus to the heart,
increaseinheartratelevelcorrespondstotheincreasesinheart changing its dynamics (Stauss et al. 1997): inspiration (expi-
rate occurring with subsequent phee call production. ration) should increase (decrease) the heart rate. If respiration
Consideringthatmammalianvocalizationsareproducedina decreases, heart rate should increase, and vice versa. Consid-
largely similar manner through the utilization of respiratory eringthisactivity,thecardiorespiratorydynamicsduringvocal
powerandlaryngealtension(FitchandHauser1995;Ghazan- production could simply be the result of the baroreflex main-
far and Rendall 2008), any shared mechanisms across species tainingcompensatorycardiacactivityafterthesubjectmodifies
are likely to include the baroreceptor pathways. The act of its breathing to produce a vocalization. Figure 3B shows that
vocalizing is critically dependent on respiration, and respira- this compensatory mechanism indeed occurs during vocal
tory output is linked to cardiac activity via baroreceptor path- production, when respiration rate significantly (red-lined ep-
ways (Critchley and Harrison 2013). We wanted to test if the ochs)decreaseswhiletheheartrateincreases.Moreover,atthe
change in respiration during vocal production could directly offset of the vocalization, when the heart rate is decelerating,
account for the changes we observed in cardiac activity 23-s the respiration rate rebounds. However, not everything can be
beforepheecallproduction(Fig.3A).Thatis,maybethevocal explained by this compensatory mechanism: For several sec-
production-relatedcardiacchangesweobservedarenotrelated onds prior to the vocalization, when heart rate is increasing,
toarousal,perse,butaretheconsequenceofrespiratorysinus there is no change in the respiration rate.
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
MAYER WAVE AND VOCAL PRODUCTION 759
To explore the extent of the RSA contributions to the changesinheartrateobservedduringpheecallproduction,we
cardiorespiratory patterns observed around a vocalization calculatedamotionindexforeachanimalusingvideorecorded
event, we tested if heart rate could be linearly predicted from during the experimental session (Fig. 4A). Figure 4B demon-
respiratory activity when a vocalization is produced. We first stratesthetraceforthemotionindex40sbeforeandaftercall
cross-correlated heart and respiratory rate signals (Fig. 3C). onset. Red lines indicate epochs that were significantly differ-
Next, we built linear predictors of heart rate from ongoing ent from baseline motor activity at the 5% significance level.
respiratory rate using both the positive and negative peak of Beginning 10 s before vocal onset, there is a gradual decrease
correlation(Fig.3C).Wethencreateda95%confidenceregion
in motor activity, with the cessation of movement reaching
ofpredictedheartrateforthedataalignedtocallonsetandcall significance (cid:2)3 s before call onset. Once the call occurs,
offset (Fig. 3D). When the median heart rate activity is com-
activity resumes. We sought to determine whether we could
pared with the predicted linear changes in heart rate activity,
predicttheobservedcardiacdynamicsusingalinearpredictor
boththeaccelerationbeforevocalonsetandthedecelerationat
analogous to the one we built for cardiorespiratory activity
vocal offset violated this prediction (exceeded a 95% confi-
(Fig. 3, C and D). We constructed predictive regions of heart
dence region). Therefore, during vocal production, respiratory
rate activity from the positive and negative peak of the cross
rate fails to predict cardiac dynamics. These results suggest
correlationbetweenmotionandheartrate(Fig.4C).Movement
that cardiac-related interoceptive processes during vocal pro- D
ductioncannotsolelybepredictedbythehomeostaticinterac- could predict the slow increase in cardiac activity but not the ow
tions between the heart and lungs. Could another indicator of deceleration after call offset (Fig. 4D). This analysis demon- n
lo
arousal account for the observed changes in heart rate? stratesthattheincreaseinheartratecanbepredictedbymotor a
d
Thelinkbetweenheartratechangesandvocalproductionis activity occurring 6–8 s before the change in dynamics. e
d
not wholly predicted by general increases in activity levels. Thoughpredictiveofheartratedynamicsbeforevocalonset,it fro
Across species, changes in general motor activity are consid- is unclear whether the motor activity is causally related, as m
ered to be a reliable indicator of arousal state (Garey et al. increases in cardiac activity during active movement typically h
2w0i0th3;chPafnagffes2i0n06h)e.arPthryasteic,awl haecrteivbiytycaisrdipaocsiaticvtievlyityciosrrheilgahteedr oScmciutrhientsatal.nt1a9n7e6o)u.sly (within a cardiac cycle) (Obrist 1981; ttp://jn
during movement than quiescence (Obrist et al. 1970). To Thus neither respiration nor motion can wholly predict the .ph
y
determinewhetherchangesinmotoractivitycouldpredictthe fluctuations in cardiac activity. How, then, do patterns of s
io
A log
y
.o
ce Calculate pixel difference for each second of video Convert into binary using percentile threshold rg
solute differenof pixel values 000...0000246 by 10.220.3/
Ab -40 -30 -20 -10Tim0e (s)10 20 30 40 -40 -30 -20 -10Tim0e (s)10 20 30 40 Fig. 4. General motor activity partially ac- 3.5 o
B counts for fluctuations in arousal during a n
vocalization. A: schematic showing how we A
u
0.2 0.2 calculatethemotionindexforeachanimal.B: g
n.u.) 0.15 n.u.) 0.15 stomroaoctthiveidtyppolpouttleadtiownitmhethdeia9n5o%fobnogootsitnrgapmpeod- ust 1
ex ( ex ( CouItsoifdeth9e5%meodfiatnh.eRbeodotslitnraeppineddicsaigtensifivcaalnucees 7, 2
nd 0.1 nd 0.1 testformotorvariability.C:cross-correlation 01
n i n i plot of the motion and heart rate signals for 6
otio 0.05 p9<5%0.0 C5I otio 0.05 doafftsaeta(lgigrnayed).Ttohecaplolsiotinvseet(p(ubrlpalcek))anadndnecgaal-l
M 0 Avg. call duration M 0 tive(green)peaksofthecrosscorrelationare
-40 -30 -20 -10 0 0 10 20 30 40 denoted with crosses and were used to con-
structalinearpredictor.D:resultsofalinear
Time to call onset (s) Time from call offset (s)
predictor of heart rate from ongoing motor
C D activity.Thesmoothedpopulationmedianof
heart rate is plotted (black) against the 95%
nt 1 Aligned to onset e) 70 e) 70 bootstrappedCIofthemotion-predictedheart
cie Aligned to offset ntil ntil r(agtreeena)tpietsak.positive (purple) and negative
oeffi 0.5 erce 60 erce 60
c p p
on 0 e ( 50 e ( 50
Correlati -0-.15 Heart rat 3400 +- pprreedd.. ssppaaccee Heart rat 3400 +- pprreedd.. ssppaaccee
0 10 20 30 40 -40 -30 -20 -10 0 0 10 20 30 40
Time offset (s) Time to call onset (s) Time from call offset (s)
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
760 MAYER WAVE AND VOCAL PRODUCTION
cardiorespiratory activity correspond the spontaneous produc- measured by heart rate (blue, left), and just before the trough,
tion of vocalizations? as measured by respiratory rate (yellow, right). For both heart
Periodic autonomic nervous system fluctuations and vocal rateandrespirationrate,theseclustersaroundparticularphases
production:theMayerwave.Itisknownthatneuralactivityin exceeded the bootstrapped 95% confidence interval (P (cid:5)
thebrainismodulatedbyrhythmicautonomicnervoussystem 0.001),andtheconcentrationsoftheclusterswerealsosignif-
activity (Critchley and Harrison 2013) and that the timing icant (Cramer-Von Mises test). The onset of a vocalization is
(phase) of external events relative to these interoceptive thus preferentially located along a particular phase of the
rhythms influence many behaviors (Barrett and Simmons Mayer wave.
2015). Such a relationship may exist between rhythmic auto- Despite producing vocalizations in an oscillatory manner,
nomic nervous system activity and vocal production. We marmosets do not produce a vocalization with every cycle of
hypothesized that the drive to produce a vocalization, in the theMayerwave.Wehypothesizethatthisbehavioralvariabil-
absenceofarelevantandexternalconspecificsignal,isphase- ity implies that the Mayer wave interacts with spontaneous
locked to rhythmic cardiorespiratory activity. vocal production in a manner similar to stochastic resonance
Figure 5A shows an exemplar of a one-syllable phee call (Fig.5E).Underamodelofstochasticresonance(Gammaitoni
from monkey 4 demonstrating a pronounced rhythmic oscilla- et al. 1998), vocal production occurs when a threshold is
tion in heart and respiratory rate. Figure 5B is the probability reached. The Mayer wave serves as a weak signal that, alone, D
o
densityoftimebetweentheonsetsofconsecutivevocalizations istypicallyunabletoreachthethresholdandproduceacall.If w
n
and shows that spontaneous vocal production is rhythmic in a source of noise adds to the weak signal, the threshold is lo
a
nature (Takahashi et al. 2013). Figure 5C shows an average breachedandvocalproductionoccurs.Inthismodel,thenoise d
e
powerspectraldensityestimateoftheperiodencompassing40 isunidentifiedbutcouldbeoneofthemanyprocessesthatlead d
sbeforeavocalizationforheartrate(blue)andrespirationrate to a vocalization (see DISCUSSION). fro
m
(yellow). There is a peak frequency of 0.07 Hz for heart rate A dynamical systems view of vocal production. Since the
h
and0.099Hzforrespiratoryrate.Thesefrequenciesfallinthe autonomic nervous system is critical for maintaining homeo- ttp
rnaonmgeicofnethrveoMusayseyrstwemavew, tihthe raesopnerainotdiscigitnyatuarroeuonfdth0e.1auHtoz- swtaeticsocuognhtrtoltoindtheetemrmaminme ahlioawnbtohdeyo(LboseerwvyedanflduScptuyaetrio1n9s90i)n, ://jn.p
((cid:2)0.04–0.15 Hz) and present in every mammalian species autonomicsignalsduringvocalproductioncorrespondtobase- h
y
studied to date (Julien 2006). Thus, in this context, the heart line arousal state dynamics. To capture this baseline, we fit a sio
and respiratory rate rhythms are both products of the Mayer separate multivariate Gaussian distribution to each pair of the lo
g
wave,therhythmicactivityofthesympatheticnervoussystem. bootstrappeddatasetfromheartandrespiratoryratevariability y
.o
We also observed a peak frequency of 0.145 Hz for phee call used in Fig. 3, A and B, and from motor activity in Fig. 4B rg
rate (Fig. 5C, gray). To determine whether there was a rela- (Z-scored medians plotted in Fig. 6A). The circular regions in b/
y
tionship between the observed Mayer wave and call produc- Fig. 6, B–D, encompass the 95% confidence interval for each 1
0
tion, we calculated the phase of respiration and heart rate distribution.Againstthisbaselineregion,weplottedingraythe .2
relative to the onset of phee calls. Figure 5D reveals that the trajectory for the median of each dynamic for the entire 2
0
vocalizationonsetoccursafterthepeakoftheMayerwave,as population. Traces of heart rate, respiration rate, and motor .3
3
.5
A B C E o
n
A
Amp.Freq.(n.u.)(kHz) 1-680101 0 1 2 M4 Probability density000...000135 Magnitude (n.u.)00001.....35791 0Weak1 0MayeT20rim wea v(3se0) sign4a0l 50tDhreecsishioolnd ugust 17, 2016
10 20 30 40 50 60 70 0.1 0.2 0.3 Noise
ee) 1 D Call interval (s) Frequency (Hz)
eart ratercentil0.5 0.4 90 90 0.4
H(p 0 0 . 30.2 0 . 20.3 0 10 T20ime (3s0) 40 50
1 180 0 180 0
ee) Mayer wave + noise
esp. ratercentil0.5 180 180 Phee call
R(p 0-20 -15 -10 -5 0 5 10 15 20 270 90 270 90 270
0 10 20 30 40 50
Time (s) 0 0 Time (s)
Fig. 5. Vocalizations are phase-locked to the rhythm of the autonomic system. A: single exemplar of a 2-syllable vocalization The exemplar comprises the
vocalization’sspectrogram,theamplitudeofthevocalization,thecalculatedheartrate,andthecalculatedrespiratoryrate.B:probabilitydensityshowingthe
oscillatory nature of spontaneous phee call production. C: power spectrum density estimates for the call rate of marmoset vocal production (gray) and for
theheart(blue)andrespiratoryrate(gold)precedingvocalonset.Shadedregionsaroundaveragesrepresent95%bootstrappedCIs.D:polarhistogramsofthephase
locationfortheslowfluctuationinheart(left,blue)andrespiratoryrate(right,gold).Heightcorrespondstothepercentageofvalueswithintherespectivebin.Translucent
histogramrepresents95%bootstrappedsignificancelevel.Atcenter,aschematicdemonstratesthedegreevaluesagainsttheirrespectivelocationsonaschematicwave.
E:astochasticresonancemodelforhowtheMayerwaveinfluencesspontaneousvocalproduction.They-axisrepresentsaneuralactivity.
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
MAYER WAVE AND VOCAL PRODUCTION 761
A B
2
25 sec. before onset Call onset
1 60
20 sec. after onset Avg. offset
Fig. 6. Vocalizations are produced when pertur-
e
e 0 at 55 bations of the system take the cardiorespiratory
scor -1 on r 50 dscyonraemdiscisgnaawlsafyorfhreoamrtraantea(bttlruaec)t,orressptairtaet.ioAn:raZte-
Z- ati (gold),andmotoractivity(red)from40sbefore
-2 pir 45 callonsetuntil20saftercallonset.Shadedregion
Heart rate s indicatestheaveragedurationofacall.B–D:state
-3 Respiration rate Re 40 spaces for the median cardiorespiratory activity
Motion index
reported in Fig. 3, A and B (B), cardiomotor
activity reported in Figs. 3A and 4B (C), and
-20 -10 0 10 20 35 40 45 50 55 60
respiratorymotoractivityreportedinFigs.3Band
Time (s) Heart Rate
4B (D). Each graph represents median activity
from 25 s before call onset until 20 s after call
C D
onset.Thecircularregionoutlinesthe95%con-
fidenceintervalsformultivariatenormalGaussian D
60 60 distribution of the bootstrapped significance test ow
for the relevant data from Figs. 3A, 3B, and 4B. n
55 e 55 lo
at Greenandredcrossesdenotethebeginningandend a
Heart rate 34455050 Respiration r 34455050 obocbaflfeaftttsecwheskeet,teahtrnrreerasotjdhpewyceenstcoaatsirrmviyrgoe,incwlryi’efss.ysTdpdcieehracleeiltcnlidvteoiieaornletnysec,e,sttaiwotnahnhdneedotrchfeateahtlhlselieegtahvhaveertenrbbrota.llwuugeeeinlacidnanilde-l httpded from
://jn
0.04 0.08 0.12 0.16 0.04 0.08 0.12 0.16 .p
Motion index Motion index hy
s
io
activity begin 25 s before call onset, continue for the duration typeofcontactcall,thepheecall,withintheundirectedcontext log
y
ofthecall,andend20saftercallonset.Figure6demonstrates of social isolation. This call is produced with some regularity .o
the state space for the median arousal state for cardiorespira- (onceevery10sorso)withouttheneedforanexternalsensory rg
tory activity (Fig. 6B), cardiomotor activity (Fig. 6C), and trigger such as the presence of, or vocalization from, a con- by/
respiratory-motoractivity(Fig.6D)aroundtheproductionofa specific. We show that heart rate influences the timing and 1
0
pheecall.Greenandredcrossesdenotethebeginningandend probability of vocal production. Gradual increases in cardiac .2
2
oarfrotwhes tsriagjneicftyorcya,llreosnpseecttiavnedlyt,hewhavereeraasgethceallbloufefseatn,drebsplaecck- apcrotidvuitcytioanre,foobllsoewrveeddbryouagrhalpyid23ansdbperfoolorengthede doencseelteroaftivooncianl 0.33
tively. The light blue line between the arrows delineates the heartrateof(cid:2)8sinduration.Thetimingofvocalproductions .5 o
call event. Surrounding the production of a phee call, we see n
was related to the phase of on-going fluctuations of cardiore- A
numerous interactions between motor activity, heart rate, and spiratory activity. These fluctuations were on the order of 0.1 ug
respiratory rate. Cardiorespiratory activity oscillates at the u
edgeofthe95%regionofthestatespace,and(cid:2)23sbeforethe HcaztoarnodfsryepmrpesaethnetetidctnheervMouasyseyrswteamvea,ctaivwiteyll(-Jeusltiaebnli2s0h0e6d).inItdiis- st 1
call, the activity leaves the baseline region as heart rate in- 7
creases, followed by a quick drop in respiratory rate (cid:2) 2 s important to note that the phee call is only one of several , 2
different calls produced by marmosets, but other types are 0
before call onset. This occurs as motor activity is gradually heard in different contexts (Bezerra and Souto 2008); our 16
diminishing. During the call, motor activity has reached its
pattern of results may not generalize to these other call types.
troughandthedirectionofthedynamicswitches.Thedynamic
Thechangesinheartratethatbrackettheproductionofphee
activity traverses across the state space, returning to the
calls by marmoset monkeys are similar to those observed
baseline.
duringhumanspeech.Humansalsoexhibitanincreaseinheart
We consider this baseline as an attractor. A vocalization is
rate before speaking (Linden 1987; Lynch et al. 1980), even
produced when the dynamics are driven far away from this
duringsignedcommunicationbydeafindividuals(Malinowet
attractor.Thus,duringthemomentsprecedingandfollowinga
al.1986).Interestingly,nonverbalutterancesinhumansdonot
phee call, the arousal state is driven well outside the expected
fitthispattern.Mirthfullaughterelicitedduringstructuredplay
range of its variability. These trajectories, therefore, depict
doesnotdemonstrateanydifferenceinheartratebeforeorafter
how interoceptive processes can influence spontaneous vocal
a bout of laughter (Fry and Savin 1988). The cardiac and
productionandtheextenttowhichavocalevent,thepheecall,
respiratorysystemsarehighlyintertwinedsuchthatthesimple
propels the state beyond its typical activity range.
actofbreathingairintothelungstemporarilyblocksparasym-
patheticinfluenceontheheart,increasingheartrate(Berntson
DISCUSSION et al. 1993). Conversely, exhaling restores parasympathetic
Weinvestigatedthearousaldynamicsrelatedtothesponta- influence on the heart, decreasing heart rate. This dynamic is
neous, uncued vocal production in marmoset monkeys. We known as respiratory sinus arrhythmia (RSA; Bernardi et al.
exploited the fact that this species readily produces a single 2001;Berntsonetal.1993),andthereisacorrelationbetween
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org
762 MAYER WAVE AND VOCAL PRODUCTION
itandheartrate(GrossmanandTaylor2007).Sinceproducing We compiled the state space in which the combination of
a vocalization requires changes in respiration, we tested specific arousal states contribute to the production of a vocal-
whether the heart rate fluctuations before and after vocal ization. The take-home message of that analysis, however, is
production were simply due to the RSA and could thus be notthatarousalcausestheproductionofavocalization,perse,
linearlypredictedbyongoingrespiration.Wefound,however, but that perturbations to the resting state space increase the
that respiration rate was insufficient in predicting both the likelihood of a call occurring. Such perturbations are in the
increase in heart rate and its deceleration surrounding the call context of many other “moving parts” of the system that are
operatingondifferenttimescales.Forexample,hormonelevels
event.
influence the production of vocalizations, and vice versa.
Changes in heart rate before vocal onset were predicted by
Increasedcortisolincreasesalarmcallinginmacaquemonkeys
general motor activity of the animal; however, a causal rela-
(Bercovitch et al. 1995). Although there is no link between
tionship is not clear, because there was a 6- to 8-s delay
cortisollevelsandthepheecallsproducedbymarmosetsinthe
betweenmotormovementandheartrateincrease.Considering
undirected context (Norcross and Newman 1999), the produc-
that active movement typically induces near instantaneous
tion of vocalizations decreases cortisol levels in this species
changes in heart rate (Obrist 1981; Smith et al. 1976), it is
(Clara et al. 2008; Cross and Rogers 2006). Producing a
unclear in the current study whether the motor activity is
vocalizationisalsometabolicallycostly,andenergylevelsare D
causally related to the heart rate changes we observed. The anotherfactorthatcaninfluencevocalproduction(Ryan1988). ow
accelerationofheartrateanddecreaseinmotoractivitybefore Moreover, simply hearing the call of another can send the n
lo
vocalonsetcouldbetheresultofanticipatoryefforttoproduce cardiorespiratory state space into fluctuation (Berntson and a
d
a call. However, in human reaction time studies, there is an Boysen 1989), and these changes in arousal level may influ- e
d
observed anticipatory deceleration in heart rate before task ence the likelihood of a vocal response and the acoustic fro
performance (Jennings et al. 1990; Obrist et al. 1969; Somsen structure of this response (Choi et al. 2015; Takahashi et al. m
et al. 2004). Functionally, a state of cardiac deceleration 2013). Thus the production of a vocalization is influenced by h
froedlluocweidngarothuesapl.roTdhuucsticoanllionfgaincothnitsacmtacnanllerimmpalyiebseaasstaanteaocft mmaonnyesfaacntodrse:ntehrogsye olecvceulrsr)inigntaetrasclotiwngerwtiimthestchaoleses (oec.gc.u,rhrionrg- ttp://jn
of self-soothing (Gracanin et al. 2014). more quickly (e.g., sensory inputs, cardiorespiratory changes, .ph
y
The oscillatory signature of the autonomic nervous system, neural activity). s
io
the Mayer wave, is present in both cardiac and respiratory In the undirected context studied here, the perturbations of lo
measures. The Mayer wave represents perturbations to the resting arousal state were represented by three consecutive gy
baroreflex (Julien 2006), resulting in an unstable negative events: initial slow increase in heart rate, followed by the .o
rg
feedbackcontrolloopthatgeneratesself-sustainedoscillations cessation of movement, which is then succeeded by a quick b/
at its resonance frequency. Most mammals exhibit a range of decrease in respiration rate. Each of these three events drives y
(cid:2)0.04to0.15Hz.Weobserveda0.145-Hzfrequencyofvocal the dynamics of the arousal system far away from its baseline 10
.2
productionacrossourmarmosetsubjectsanddemonstratedthat region, resulting in vocal production. In dynamical systems 2
0
vocal onset occurs at a preferred phase of the ongoing Mayer terminology, the baseline region is an attractor and the vocal- .3
wave. Yet, a vocalization was not produced in every cycle of ization is produced when a perturbation takes the dynamics 3.5
the Mayer wave, and this suggests a mechanism akin to away from the attractor. In this scenario, we hypothesize that o
n
stochastic resonance (Fig. 5E). Stochastic resonance requires the vocalization is a homeostatic mechanism to return the A
threebasicingredients:1)athreshold,2)aweakcoherentinput arousaldynamicstothebasinofattraction,andfinallyintothe ug
such as a periodic signal, and 3) a source of noise that is attractor region. Thus one could consider vocal production us
inherent in the system (Gammaitoni et al. 1998). In our itself as a form of allostatic control (Berntson and Cacioppo t 1
7
scenario, the Mayer wave is the periodic signal that, with the 2007; McEwen and Wingfield 2003) or one parameter of a , 2
addition of as-yet undefined source(s) of input, leads to a cybernetic system whereby compensatory actions (vocal pro- 0
1
breachinthedecisionthreshold,producingavocalizationwith duction, in our case) offset the effects of disturbances (drifts 6
someprobabilitygreaterthanzero.Inscenariosinvolvingmore away from the resting arousal state) (Pellis and Bell 2011).
than one individual, the input signal would include the facial The neural pathways for these putative interactions involv-
and/orvocalexpressionsofconspecifics(GhazanfarandTaka- ingarousalandvocalproductionoverlaptoasurprisingdegree
hashi 2014). Considering the relationship between vocal onset and provide a viable pathway by which arousal influences
and the ongoing phase of the Mayer wave, an indirect conse- vocalproduction.Atthecorticallevel,theinteroceptivesystem
quenceofthisslow,pervasiveautonomicoscillationistoassistin consists of a network of areas along the medial wall of the
theassemblyofvocalbehavior.Indeed,vocaldevelopmentaldata cerebral cortex, from the cingulate cortex to the ventromedial
from marmosets suggests that the proper assembly of contact prefrontal cortex, as well as laterally into the orbitofrontal
calling occurs postnatally, with an upward shift of the threshold cortex and insula (Barrett and Simmons 2015; Critchley and
(ZhangandGhazanfar2016).Infantmarmosetsintheundirected Harrison 2013). These visceromotor cortices generate auto-
contextproducevocalizationsatalmost10timestherateofadults, nomic,hormonal,andimmunologicalpredictionstoadjusthow
with a range of immature-sounding call types. After about a theinternalsystemsofthebodydeployresourcestodealwith
month, they slow the rate of call production and produce phee the external world in the immediate future (Barrett and Sim-
calls almost exclusively. The speed with which this transition mons 2015). Collectively, they project subcortically to the
fromimmaturetomaturepheecallstakesplaceisinfluencedby amygdala, ventral striatum, hypothalamus, and the periaque-
contingent vocal feedback provided by parents (Takahashi et al. ductal grey (An et al. 1998; Barrett and Simmons 2015;
2015,2016). CritchleyandHarrison2013).Thisnetworkorganizationover-
JNeurophysiol•doi:10.1152/jn.00136.2016•www.jn.org