1: 

2: %

3: \documentclass[aps,twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

4: %

5: %

6: %

7: \usepackage{colordvi}

8: %

9: %

10: \input{epsf}

11: 

12: \def\bc{\begin{center}}

13: \def\ec{\end{center}}

14: 

15: 

16: 

17: \def\be{\begin{equation}}

18: \def\ee{\end{equation}}

19: 

20:  \def\bi{\begin{itemize}}

21: \def\ei{\end{itemize}}

22: 

23: \def\bea{\begin{eqnarray}}

24: \def\eea{\end{eqnarray}}

25: 

26: \def\bR{{\sf R}}

27:  \def\bv{{\bf v}}

28:  \def\br{{\bf r}}

29:  \def\bs{{\bf s}}

30:   \def\bV{{\bf V}}

31:  \def\bE{{\bf E}}

32: \def\bL{{\bf L}}

33: \def\bC{{\bf C}}

34: \def\bK{{\sf K}}

35: \def\bQ{{\sf Q}}

36: \def\bA{{\sf A}}

37: \def\bB{{\sf B}}

38: \def\bW{{\sf W}}

39: \def\bJ{{\sf J}}

40: \def\bH{{\sf H}}

41: \def\bT{{\sf T}}

42:  \def\i{{\it i}}

43:  \def\ri{{\rm i}}

44: \def\cc{\it c.c.}

45:   \def\Re {{\it Re}}

46:   \def\Im{{\it Im}}

47:   \def\bI{{\bf I}}

48: \def\bJall{{\cal J}}

49: \def\bWall{{\cal W}}

50: \def\cH{{\cal H}}

51: 

52: \def\wasred{}

53: \def\wasblue{}

54: 

55: 

56: \begin{document}

57: \title{Mathematical analysis and simulations of the neural circuit for

58: %

59:  locomotion in lamprey}

60: \author{Li Zhaoping$^1$, Alex Lewis$^1$, and Silvia Scarpetta$^2$}

61: \affiliation{$^1$University College, London, UK\\

62: $^2$INFM \& Dept.of Physics, University of Salerno, Italy}

63: \begin{abstract}

64: %

65: %

66: %

67: %

68: %

69: %

70: %

71: %

72: %

73: %

74: %

75: %

76: %

77: %

78: %

79: %

80:  We analyze  the dynamics of the neural circuit of the

81: {lamprey} central pattern generator (CPG) 

82: %

83: This analysis provides insights into how neural interactions 

84: form oscillators and enable spontaneous oscillations in a network 

85: of damped oscillators, which were not apparent in previous simulations

86: or abstract phase oscillator models.

87: We also show how  the different behaviour regimes (characterized by phase and amplitude 

88: relationships between oscillators) of forward/backward swimming,

89: and turning, can be controlled using the neural connection

90: strengths and external inputs. 

91: 

92: \pacs{{87.19.La, 87.19.St}}

93: \end{abstract}

94: \maketitle

95: 

96: %

97:  \wasred{Locomotion in verbebrates {(walking}, 

98:  swimming, etc.) is}

99: generated by  central pattern generators (CPGs) in the spinal cord.

100: The CPG for  {swimming in lamprey}

101: is  \wasred{one of the best known

102: \cite{Grillner2002,Buchanan2001}}, and

103: has been a model system for 

104: %

105:  \wasred{investigations.}

106: {It}

107: produces left-right {anti-phase} oscillatory neural and motor activities

108: propagating along a body composed of around 100 segments. 

109: \wasred{A head-to-tail negative or positive oscillation phase

110:   gradient,

111:   of about

112: 1\% of an oscillation cycle per segment, gives forward or background 

113: swimming respectively, and

114: one wavelength from head to tail.}

115: %

116: %

117: %

118: %

119:  \wasred{External inputs from the brain stem {}{switch} the CPG between 

120: %

121: } forward and backward swimming of

122: various speeds and turning.

123: Since {isolated sections} of the spinal cord,

124: \wasred{ down to 2-3 segments long \cite{Buchanan2001}},

125: can produce  swimming-like {activity}, 

126: %

127: the

128: oscillations are thought to be generated by the neurons

129: within the  \wasred{CPG}.

130: %

131: %

132: The neural circuit responsible is shown topologically

133: in Fig. \ref {fig:circuit}. It has

134:  \wasred{ipsilaterally projecting 

135: %

136: excitatory ($\bE$) neurons and 

137: inhibitory ($\bL$) neurons,

138: and contralaterally projecting 

139: %

140:  inhibitory $(\bC$) neurons, and

141: provides output to motor neurons via the $\bE$ neurons.}

142: %

143: {All {neurons} project  both intra- and inter-segmentally.

144: %

145: %

146:  \wasred{The projection distances are {}{mainly} within a few

147: segments, especially from $\bE$ and $\bC$ neurons, and  

148: are longer, and possibly stronger, in the head-to-tail 

149: or descending direction} \cite{Grillner2002}-\cite{McClellan1998}. 

150: \begin{figure}[t]

151: \setlength{\unitlength}{0.6pt}

152: \begin{center}

153: %

154: %

155: %

156: \mbox{\epsfxsize=250pt \epsfbox{fig1.eps}}

157: \end{center}

158: \caption{\label {fig:circuit} The lamprey CPG circuit. The solid and dashed lines

159: denote intra- and inter-segment connections respectively. }

160: \end{figure}

161: %

162: 

163: %

164: %

165:  Previous analytical work \cite{CohenEtal82, E&K}

166: {mainly} treated the CPG as a chain of coupled phase oscillators 

167:  \wasred{in a general form $\dot\theta_i = \omega_i +

168:  \sum_jf_{ij}(\theta_i,\theta_j)$.

169:  Here $\theta_i$ is oscillation phase and $\omega_i$

170:  {is}  intrinsic frequency,

171: modelling the  

172: behaviour of  {one} segment, and

173: $f_{ij}(\theta_i,\theta_j)$

174: {models} inter-segmental coupling.}

175: This approach  \wasred{provided important insights into

176: the conditions {for}

177: %

178:   phase-locked solutions

179: applicable} to  \wasred{various} systems of coupled oscillators. 

180: However, 

181: %

182:  \wasred{its generality} obscures the

183:  %

184:  {roles of specific} neural types and their 

185: \wasred{connections}

186: in generating and controlling behaviour.

187: \wasblue{More recently, %

188: {bifurcation} analysis of the dynamics

189:   of a single segment was carried out,} for a

190:   phase oscillator model derived from a kinetic (Hodgkin Huxley)

191:   equation for neurons \cite{Taylor1998}, and   

192:   for a neural circuit model similar to the one used in this paper

193:   \cite{Jung}.}

194:  {Extensive simulations}, including all neural types

195:    {}{and detailed neural}  properties,

196: have reproduced many features of experimental data \cite{Grillner2002},

197: though the {model's} complexity limits further

198: understanding.

199: 

200:   

201: In all previous approaches, it is assumed that a single segment in the

202: {}{CPG can oscillate} spontaneously, contrary to experimental evidence

203: that at least 2-3 segments are needed for

204: oscillations \cite{Buchanan2001}.

205: We present  \wasred{an} analytical study, confirmed by simulations, of 

206: a model of the CPG neural circuit in which {}{an isolated} single

207: segment

208: has a stable

209: fixed point, with spontaneous oscillations occurring only  in 

210: chains of coupled segments.

211: The phase oscillator approach is not

212: applicable here since it assumes spontaneously {{oscillating 

213: individual segments perturbed by inter-segment coupling}.

214: Including specific cell types and their {connections} enables us to

215: analyse the role of each of them in generating and controlling

216: swimming.

217: We 

218: show how external inputs select forward

219: and backward swimming, by {}{controlling

220: the relative strengths of connections 

221: between various neurons}, and produces 

222: turning,  by additional input to one

223: side of the {CPG} only.

224: We also {analyse}  behaviour near the {body ends}. 

225:  

226: 

227: We model the CPG neural circuit 

228: %

229:  \wasred{with} $N = 100$ segments denoted by {$i=1,

230: .,N$.} 

231: The vector states 

232:  \wasred{($\bE_l,\bL_l,\bC_l$) and

233:  \wasred{$(\bE_r,\bL_r,\bC_r)$}, modelling the membrane potentials of the local populations of neurons  \wasred{at the left and

234: right side of the body respectively},

235: with $\bE \wasred{_l} = (E^1 \wasred{_l},E^2 \wasred{_l},\cdots,E^N \wasred{_l})$ etc.,

236: are modelled as leaky integrators of their inputs:}

237:  \bea

238: \dot \bE_{ \wasred{l}} &=& - \bE_{ \wasred{l}} - \bK^0 g_C(\bC_{ \wasred{r}}) + \bJ^0

239: g_E(\bE_{ \wasred{l}}) +\bI_{E, \wasred{l}}

240: \nonumber \\

241: \dot \bL_{ \wasred{l}} &=& - \bL_{ \wasred{l}} - \bA^0 g_C(\bC_{ \wasred{r}}) + \bW^0

242: g_E(\bE_{ \wasred{l}}) +\bI_{L, \wasred{l}}

243: {\label {uno}}\\

244:  \dot \bC_{ \wasred{l}} &=& - \bC_{ \wasred{l}} - \bB^0 g_C(\bC_{ \wasred{r}}) + \bQ^0 g_E(\bE_{ \wasred{l}}) -

245: \bH^0 g_L(\bL_{ \wasred{l}}) +\bI_{C, \wasred{l}} \nonumber 

246:  \eea

247:  with the same equation for 

248:  swapped subscripts 

249:  {$(l\leftrightarrow

250:   r)$.}

251: {$g_E(\bE_l) = (g_E(E^1_l),\dots,g_E(E^N_l))$}

252: are the neural activities or firing rates, 

253: %

254:  \wasred{as non-negative (sigmoid-like) activation functions of $\bE_l$,}

255: and likewise for  \wasred{$g_C(\bC_l)$ and $g_L(\bL_l)$}. $\bK^0$,

256: $\bJ^0$, 

257: $\bA^0$, $\bW^0$, $\bB^0$, $\bQ^0$, and $\bH^0$ 

258: are 

259: $N\times N$

260: matrices  \wasred{of} non-negative elements

261: %

262:  \wasred{modeling} the synaptic strengths between neurons.

263: %

264: %

265: $\bI_{E, \wasred{l}}$, $\bI_{L, \wasred{l}}$, and $\bI_{C, \wasred{l}}$ are

266: external inputs, \wasred{including those from the  brain stem},

267: assumed to be  {static}. 

268: %

269:  \wasred{A left-right symmetric

270:   fixed point $(\bar\bE,\bar\bL,\bar\bC)$ where

271: $(\dot\bE,\dot\bL,\dot\bC)=0$ exists by setting

272: external inputs to $\bI_{E,l} = \bar \bE_l + \bK^0g_C(\bar \bC_{r}) 

273: -  \bJ^0 g_E(\bar \bE_{l})$ (and analogously for other $\bI$'s).

274: %

275: %

276: Dynamics for small deviations from $(\bar\bE,\bar\bL,\bar\bC)$

277: can be approximated linearly,

278: {and,} 

279: %

280: with a coordinate rotation

281: {$(\bE_\pm,\bL_\pm,\bC_\pm) {\equiv }

282: [(\bE_l,\bL_l,\bC_l) -(\bar\bE,\bar\bL,\bar\bC)]

283: \pm

284: [(\bE_r,\bL_r,\bC_r)-(\bar\bE,\bar\bL,\bar\bC)]$,}

285: transformed into two decoupled modes --

286: the left-right synchronous mode

287: $(\bE_+, \bL_+, \bC_+)$ 

288: and the antiphase mode $(\bE_-, \bL_-, \bC_-)$}.

289: Swimming requires oscillations, with wavelength of one body length, in

290: the anti-phase mode while the synchronous mode is damped. The

291: linearised equations are

292: \bea

293: \dot \bE_\pm &=& - \bE_\pm \mp \bK \bC_\pm + \bJ \bE_\pm \nonumber \\

294: \dot \bL_\pm &=& -\bL_\pm \mp \bA \bC_\pm + \bW \bE_\pm  \nonumber \\

295: \dot \bC_\pm &=& - \bC_\pm \mp \bB \bC_\pm + \bQ\bE_\pm -\bH

296: \bL_\pm \label{plus_minus} \eea 

297: where $\bK \equiv \bK^0 g'_C(\bar \bC )$, 

298: $\bA \equiv \bA^0 g'_C(\bar \bC )$, $\bB \equiv \bB^0 g'_C(\bar \bC )$, 

299: $\bJ \equiv \bJ^0 g'_E(\bar \bE )$, 

300: $\bW \equiv \bW^0 g'_E(\bar \bE )$, $\bQ \equiv \bQ^0 g'_E(\bar \bE )$, and

301: $\bH \equiv \bH^0 g'_L(\bar \bL)$  

302:  are effective connection matrices,

303: \wasred{and

304:   the $g'(.)$'s denote derivatives}. 

305: The $\bC$ neurons thus become effectively excitatory in the 

306:  \wasred{anti-phase mode.}

307: Noting that the lengths of the neural connections are much

308: shorter {than} 

309: %

310: the body, 

311:  \wasred{and that 

312: isolated sections of spinal cord from any part 

313: of the body generate oscillations with similar amplitude 

314: and phase relationships \cite{Grillner2002,Buchanan2001}},

315: we make the approximation of translation invariance,

316: so that matrix elements such as $\bJ_{ij}$

317: %

318: depend only on ($i-j$),

319: and impose the periodic boundary condition, 

320: %

321: {$\bJ_{ij}= \bJ(x)$,} where $x= (i-j) \bmod N$. This is adequate 

322: \wasred{when

323: behaviour near body ends is not considered.}

324: Then all connection matrices {}{commute with each other},

325: with common eigenvectors (expressed as functions of segment number $x$)

326:  \wasred{$(\bE(x), \bL(x), \bC(x)) \propto e^{i (2\pi m/N)x}$ for

327: {integer} eigenmode $-N/2< m \le N/2$.

328: The system solutions are thus combinations of modes

329:  $(\bE_\pm (x,t), \bL_\pm (x,t), \bC_\pm (x,t)) \propto

330: e^{\lambda^\pm _m t + i (2\pi m/N)x}$ where $\lambda^\pm _m$ is

331: eigenvalue of eq. (\ref{plus_minus})

332: for mode $m$. Forward swimming results if 

333: the real part $\Re(\lambda^\pm _m ) < 0$ for all modes except

334: the antiphase mode with $m=1$, ie $Re(\lambda_1^-)>0$. Then this

335: mode dominates the 

336: solution (whose growing amplitude

337: will be constrained by nonlinearity)

338: $(\bE (x,t), \bL (x,t), \bC (x,t))

339: \propto e^{ \Re (\lambda^-_1 ) t -i (\omega t - kx)}$,

340: with oscillation frequency $\omega \equiv \left|\Im (\lambda^-_1)\right|$ and

341: wave number $k = 2\pi /N$.

342: Using the convention $e^{-i\omega t}$ for oscillations, we omitted

343: the solution $\propto e^{\Re(\lambda^-_1 ) t +i \omega t}$ in 

344: the conjugate pair of eigenvalues.}

345: %

346: %

347: %

348: %

349: %

350: %

351: %

352: To simplify our system, {we} note from experimental data that in forward swimming,

353: {$\bE$} and {$\bL$} oscillate roughly in phase within {a}

354: segment, while  {$\bC$} 

355:  leads them

356: %

357:  \cite{Buchanan2001}.

358: We scale our  \wasred{variable definitions}  {so that $\bE_-=\bL_-$

359: in forward swimming. Then } 

360:  {eq. (\ref

361: {plus_minus})} implies 

362: %

363: %

364: %

365: that 

366: %

367: $(\bK -\bA ) \bC_-  = -(\bJ - \bW ) \bE_-$ in forward swimming.

368:  \wasred{Since $\bE$ and $\bC$ have much shorter connections than

369:   the  wavelength of oscillations during swimming, the connection

370:   matrices have zero elements far from the diagonal, making

371:   $(\bK -\bA ) \bC_-$ and $(\bJ - \bW ) \bE_-$  roughly either in

372:   phase or in

373:   anti-phase 

374: with $\bC_-$ and $\bE_-$ respectively.

375: As $\bC_-$ phase leads $\bE_-$, $(\bK -\bA ) \bC_-  = -(\bJ - \bW ) \bE_-$

376: is impossible unless $(\bJ -\bW)\bE_-=(\bK - \bA)\bC_-=0$.}

377: For simplicity we henceforth assume $\bJ =\bW$ and $\bK =

378: \bA$, since non-swimming modes do not concern us. Consequently

379: $\bE_\pm=\bL_\pm$ and 

380: %

381: \be

382: \begin{pmatrix}

383: \dot\bE_\pm \\

384: \dot\bC_\pm \\

385: \end{pmatrix}

386: =

387: \begin{pmatrix}

388: \bJ-1 &~~& \mp\bK \\

389: -(\bH-\bQ) &~~& \mp\bB-1

390: \end{pmatrix}

391: \begin{pmatrix}

392: \bE_\pm \\

393: \bC_\pm\\

394: \end{pmatrix}

395: \label {eqv} \ee 

396: \wasred{where $\bL$ and $\bE$  {are} treated as a single population 

397: inhibiting or exciting $\bC$ via connections $\bH-\bQ$.}

398: %

399: {The eigenvalues for mode $m$ are}

400:  {

401: \begin{align}

402: %

403: \lambda^+_m &= \left[-2 + J_m - B_m  \pm \sqrt{R_m + 2(B_m^2+J_m^2)}\right]/2 

404: \nonumber \\

405: \lambda^-_m &= 

406: \left[-2 + J_m + B_m  - i \sqrt{R_m }\right]/2.

407: %

408: \label {lams}

409: \end{align}}

410:  {$J_m \equiv \sum_x \bJ (x) e^{ -\ri (2\pi m/N) x} $} is the 

411: eigenvalue of $\bJ$ (and analogously for other matrices),

412: and {}{$R_m$} is the eigenvalue  {of} 

413: %

414: $\bR\equiv 4\bK  (\bH-\bQ) - (\bB-\bJ)^2$. 

415: %

416: %

417: %

418: %

419:                                 %

420: %

421: %

422: %

423: 

424: %

425: %

426: %

427: %

428: 

429: To elucidate {}{the}

430: conditions {}{needed} for {}{the}  {antiphase} 

431: %

432: {mode with}

433: $m=\pm 1$ 

434: for forward or backward swimming to dominate, we {}{analyse the

435:   bifurcations which occur

436:   as $\lambda^\pm_m$ for each mode ($m$, $\pm$)

437: changes as the effective neural connections are varied, either directly or via

438: the external inputs.}

439: First, we focus on the left-right mode space 

440:  {(as in \cite{Taylor1998,Jung} 

441: %

442: for a single segment)}

443: of $+$ and $-$, i.e., the synchronous 

444: and antiphase modes,  by simply taking $m =0$.

445: Then, $J_0$, $B_0$, 

446: $H_0$, $K_0$, and $Q_0$, 

447:  are all real and

448: non-negative, each {}{being} the total connection

449: strength on a postsynaptic cell from all cells of a particular

450: type. 

451: Oscillation in the antiphase mode requires $R_0>0$, necessitating 

452: $H_0>Q_0$, or that in the AC component of interactions above the background

453: DC level, $\bC$ neurons receive stronger inhibition from $\bL$ neurons 

454: than excitation from  $\bE$ neurons. 

455: Consequently, $\lambda^+_0$ is real and

456: the synchronous mode is non-oscillatory.

457: As neural connections increase from zero,

458: the  {antiphase} 

459: %

460: mode undergoes a Hopf bifurcation 

461: when  {$\Re (\lambda^-_0)=0$}, 

462: %

463: at {}{$J_0+B_0=2$},

464: {}{and} the  {synchronous} 

465: %

466: mode undergoes a 

467: pitchfork bifurcation when {}{$\lambda^+_0=0$, which occurs when

468: $(B_0+1)(1-J_0) = K_0(H_0 - Q_0)$}.

469:  {Oscillations result} if

470:   the Hopf bifurcation has occurred but the

471:   pitchfork bifurcation has not,

472:   i.e., $Re(\lambda^-_0)>0> \lambda^+_0$. The condition

473:   $\lambda^+_0<0$ implies

474:   $(B_0+1)(1-J_0) > K_0(H_0

475:   - Q_0)$, necessitating $J_0<1$.   Meanwhile,

476:   $Re(\lambda^-_0)>\lambda^+_0$

477:   leads to

478:   $B_0 > \sqrt {J_0^2 +R_0/2}>J_0$, meaning that there must be

479:   sufficient  {inhibitory}

480: connections between left and right $\bC$ cells.

481: %

482: %

483: %

484: %

485: %

486: The  $\bJ$, $\bW$, and

487: $\bQ$  {connections from} $\bE$ cells have to be relatively

488:   weak,

489:   consistent with the findings  {of \cite{Jung}}. 

490:   %

491: (If $R_0 < 0$, the antiphase mode will undergo {}{a pitchfork}

492: bifurcation, {}{and the synchronous  {mode} 

493:   either a pitchfork or Hopf bifurcation}.

494:  These {}{regimes}  are

495:  less relevant to modeling the  {lamprey}.)

496:  %

497: 

498: Assuming the synchronous mode {}{is  damped},

499: we focus now on  {the} antiphase mode in the $m$ mode space.

500: {}{Hopf bifurcations occur} sequentially in various  modes $m$ 

501: in the order of descending $\Re (\lambda^-_m)$. 

502: Taylor expanding 

503: $J_m$ (and similarly  {$B_m$, $R_m$}) for small 

504: wave number $k = 2\pi m/N$ {}{as is relevant

505: for  {swimming}, 

506: %

507: $J_m = j_0 -i k j_1 - k^2 j_2 + {\mathcal{O} }

508: (k^3)$

509: with $ j_n = \sum_x \bJ(x)\frac{x^n}{n!}$}, we have

510:   \bea

511: \!2\!\Re {(\!\lambda^-(k))\!} &=& \! - \!2 \! +\! j_0 \! +\!\! b_0\!\!-\!\! k

512: r_1/(2\sqrt{r_0})\! -\!\! k^2\!

513: (j_2\!\!+\!b_2)\!\! + \!{\mathcal{O}}\!(k^3\!) \nonumber \, \\

514: \!2\!\Im {(\!\lambda^-(k))}\! &=& \sqrt{r_0} + {\mathcal{O} } (k) 

515: \eea

516: making the mode with

517: {}{$k \approx -r_1/[4\sqrt{r_0} (j_2+b_2)] $}, {}{which has the largest

518:  $Re(\lambda^-(k))$,} dominant.

519: From the definition, {}{$(r_0,j_2, b_2) \ge 0$}, {}{while}

520: stronger and/or longer connections  in the descending

521: direction {}{imply $(j_1,b_1)>0$}.

522: Simply, $\bJ$ (and similarly for other matrices) 

523: is said to be descending (or ascending), if {}{$j_1>0$ (or  $j_1<0$).}

524: Hence, {}{if} $\bR$ {}{is ascending}, i.e., {}{$r_1<0$},

525: $\Re ({\lambda^- }(k))$ {}{increases with} increasing $k$, and the

526: dominant wave number can be set to $k=2\pi/{N}$ for mode $m=1$

527: by {tuning} the values of $\bR$, $\bJ$, and $\bB$.

528: If the connection strengths are such that

529: only the $m=1$ mode undergoes the Hopf bifurcation, 

530: forward swimming emerges spontaneously. 

531: Switching $\bR$ to descending  {leads to} backward swimming.

532: Note that $\bJ$, $\bB$, $\bH$, $\bK$, and $\bQ$ are all descending, 

533:   {multiplications} and summations of {}{descending}  connections 

534: are still {}{descending},  {and} negating a descending

535: connection makes 

536: it ascending. 

537: Since $\bB$ and $\bH$ have to dominate $\bJ$ and $\bQ$ respectively,

538: $\bR$ is composed of an ascending term $-(\bB-\bJ)^2$ 

539: and a descending term $4\bK(\bH-\bQ)$. 

540: {}{Depending on} the relative strengths of {}{these two terms}, 

541: $\bR $ can be {}{made} ascending or descending to achieve forward

542: or backward swimming.

543: This could be achieved by changing the

544:  {static inputs}

545: %

546: to shift the fixed point $(\bar{\bE},\bar{\bL},\bar{\bC})$ of the 

547: system to a different gain regime 

548: $g'_E(\bar \bE), g'_L(\bar \bL), g'_C(\bar \bC)$, and  

549: thus different effective connection strengths $\bH =

550: \bH^0g'_L(\bar \bL)$, etc. without

551: changing the underlying connection structure $\bH^0$.

552: Alternatively, the  \wasred{external} inputs {might} recruit 

553: %

554: %

555:  {extra functional cells} to alter the  {effective} 

556: %

557: connection strengths 

558: \cite{Kozlov2002}.

559: 

560: When connections are such that additional modes 

561: satisfy $\Re (\lambda^-_m)>0$, the {}{resulting behavior 

562:   depends on the nonlinear coupling between modes}.

563: For illustration, consider nonlinearity

564: only in $g_C(\bC)$.

565:  \bea

566: \dot \bE_{\pm} &=& - \bE_{\pm} \mp  \bK^0 g_\pm(\bC) + \bJ \bE_{\pm} \nonumber \\

567: \dot \bL_{\pm} &=& -\bL_{\pm} \mp  \bA^0 g_\pm(\bC) + \bW \bE_{\pm}  \nonumber \\

568: \dot \bC_{\pm} &=& - \bC_{\pm} \mp  \bB^0 g_\pm(\bC) +

569: \bK'\bE_{\pm} -\bH \bL_{\pm} 

570:  \eea 

571: where $g_\pm(\bC) = g_C(\bC_l) \pm

572: g_C(\bC_r)$. If the nonlinearity is of the form $g_C(x) = x + ax^2 - b

573: x^3 + \mathcal{O}(x^4)$, {we have} 

574: %

575:   {\begin{align}

576: g_-(C) &\approx  C_- + a C_+C_- -

577: bC_-^3/4 -

578: 3bC_-C_+^2/4  \\

579: g_+(C) &\approx  C_+ + a C_+^2/2 +

580: aC_-^2/2 -  bC_+^3/4 - 3bC_+C_-^2/4 

581: \nonumber \end{align}}

582: Hence, when $C_- = 0$,

583: $C_+$ cannot excite it since $g_-(\bC) = 0$. 

584: {However}, if $a\ne 0$, the  \wasred{synchronous mode}, 

585: will be excited  {passively} 

586: %

587: by the  \wasred{antiphase} mode through the quadratic coupling 

588: term $aC_-^2/2$, responding with double frequency, as could be easily tested.

589: 

590: To analyse coupling between {}{the}

591: antiphase modes, we {assume for simplicity that $g_c(\bC)$ is odd, so

592:   $\bC_+=0$ since the synchronous mode is damped,

593:   and  $g_-(C_-)=2g_C(C_-/2)$.}

594: Consider a small perturbation, in the $m'$ mode direction,

595: to the $m=1$ cycle (the final orbit resulting from a

596: small deviation from the fixed point in the $m=1$ mode, with

597: a fundamental harmonic in the $m=1$ mode)

598: %

599:  {such} that

600: $\bC_-(x) \approx C_1 \cos (2\pi x/N) + C_{m'} \cos(2\pi m' x/N)$ 

601: with $C_{m'} \ll C_1$.  Expressing 

602: $g_-(\bC)$ as $g_-(\bC) = \sum_n g_n e^{\ri 2\pi n x/N}$, it can be

603: shown that {for

604:   large $N$},

605: $g_{m'} \approx C_{m'} \bar g'_C$ where $\bar g'_C$ is the derivative

606: of $g_C$ averaged over the unperturbed cycle.

607: (More

608: detailed analysis will be given in a future paper.)

609:  {Because of the sigmoid form of $g_C(C)$, $\bar g'_C<g'_C(\bar \bC)$.}

610: Then $C_{m'} \propto e^{\rm\lambda^-_{m'}t}$ 

611: with $\lambda^-_{m'}$ as in equation (\ref {lams}) except that 

612: $(B_{m'}, K_{m'})$, values derived from connections from $\bC$ 

613: cells, are rescaled by a factor  {$\bar g'_C /g'_C(\bar \bC)<1$.}

614: %

615: %

616: {}{Thus the swimming cycle at large amplitude always

617:   remains stable against

618:   perturbation in other modes, even when the fixed point is unstable

619:   against these perturbations}.

620: 

621: {}{

622:   If $Re(\lambda_1^-) \gg Re(\lambda_{m'}^-) \gtrsim 0$,

623:   the $m'$ cycle  will have a small amplitude

624: and hence $\bar g'(C) \sim g'(\bar{C})$ and it will be unstable

625: against perturbation in the $m=1$ mode. For larger

626: $Re(\lambda_{m'}^-)$

627: the amplitude of the cycle is larger and either cycle will be stable.

628: } Suppose  the neural connections are such that {}{ the $m=\pm 1$

629:   cycles, giving forward or backward swimming,  are

630:   both stable}. {}{The system would then display hysteresis, with

631:   the final behaviour depending on the initial conditions.} 

632: {}{Forward or backward swimming could then be selected by}  

633: transient  inputs from the brain stem,

634: rather than by 

635: %

636:  {setting constant inputs}

637:   as described above.

638: This {}{seems less likely to be} the actual selection mechanism since

639: experiments on fictive swimming  

640: %

641: (presumably with random initial conditions) 

642: seldom observe  {spontaneous} backward swimming. {}{However, the

643:   forward swimming could simply have

644:   a larger basin of attraction than backward

645:   swimming.} 

646: 

647: When the lamprey turns,  neural activities  {on left and right sides are

648: unequal.} This is realizable by adding an additional

649: constant  {input}  to one side in the animal and

650: in models \cite{McClellanHagevik1997, Kozlov2002}, leading to {}{unequal

651: mean}  {activities} 

652: %

653: without

654: disrupting the oscillations, 

655: provided that the gains $g'(.)$ are

656: roughly constant  \wasred{near} the fixed points.

657:  {}{Simulations (Fig.2) confirm the

658: analysis above.}

659: 

660: 

661: 

662: 

663: \begin{figure}[!!!hhhhhhhh!]

664: \setlength{\unitlength}{1.0pt}

665: \begin{center}

666: \begin{picture}(350, 170)

667: \put(0,80){{\epsfxsize=240pt \epsfbox{fig2abc.eps}}}

668: \put(0,-15){{\epsfxsize=240pt \epsfbox{fig2def.eps}}}

669: \end{picture}

670: %

671: %

672: \end{center}

673: \caption{Simulations. A: Membrane potentials of $\bE$

674: population on  {either} side of one segment during forward

675: swimming and turning. The oscillations are in anti-phase between

676: the two sides. Turning is induced by an additional constant input

677: to  one side only,  {starting} at the time indicated by the dashed line. B: 

678: $\bC$ slightly phase leads $\bE$ during forward swimming. 

679: C \& D:  {Waveform} of $\bE$ along the body 

680: in forward and backward swimming, at consecutive times increasing 

681: in the direction indicated

682: by the arrows, in the translational invariant model.

683: The switch from forward to backward swimming is achieved by increasing 

684: the strength of $\bH$ and $\bQ$.

685: E: Oscillation waveforms (note different amplitudes) in body

686: segments at head, tail and centre of the body,  {without} 

687: %

688: translational invariance. F: Forward and backward swimming from

689:   different initial conditions, with the same connection strengths and inputs.

690: }

691: \end{figure}

692: 

693: To study behaviour  {at the body ends}

694: %

695: or in {}{short sections of spinal cord \cite{Grillner2002}}, or

696: equivalently to see the effects of longer connections,

697: we abandon 

698: {}{translation invariance}.

699: %

700:   {}{Eliminating 

701: $\bC$ in eq. (\ref {eqv})}, 

702: the minus mode has:

703:  \be

704:  \ddot \bE + (2-\bJ -\bB) \dot \bE + [1-\bJ - \bB +\bB\bJ + \bK(

705: \bH-\bQ)]\bE = 0, \nonumber

706:  \ee

707: or, oscillator $i$ is driven by  \wasred{force $F_i$} from {other} oscillators

708: %

709: \be

710: \begin{split}

711: \ddot \bE_i &+ (2-\bJ_{ii}-\bB_{ii}) \dot\bE_{i} + [1-

712: \tilde\bR_{ii}] \bE_{i} 

713: \label {osci_couple} \\

714:         &= F_i \equiv \textstyle{\sum_{j \ne i} F_{ij} \equiv \sum_{j \ne i}

715: (\bJ_{ij} +

716: \bB_{ij}) \dot \bE_j +\sum_{j\ne i} \tilde\bR_{ij} \bE_j}

717:  \end{split}\nonumber\ee 

718: where $\tilde\bR = \bB+\bJ -\bB\bJ

719: -\bK(\bH-\bQ)$. 

720:  {}{The intrinsic oscillation},

721: $\bE_i \sim e^{\lambda t}$, 

722: {}{is damped, $\Re(\lambda ) = -1 +(\bJ+\bB)_{ii}/2

723:  <0$},

724: as 

725: indicated by  {experiments} \cite{Grillner2002,Buchanan2001}.

726: %

727: %

728: We estimate $F_i$ 

729: %

730:  {using} the approximation that oscillators $j\ne i$ still

731: {behave as}  $\bE_j \propto e^{-{\rm i }(\omega t -kj)}$. 

732: {}{We then have} 

733: $F_i = \alpha_i  \dot \bE_i + \beta _i\bE_i$, where

734: \begin{align}

735: \alpha_i &\equiv 

736:  \textstyle{\sum_{j\ne i} \left[(\bJ + \bB)_{ij}  \cos (k(j-i))

737:   -\tilde\bR_{ij} \sin (k(j-i)) /\omega \right]}

738:  \nonumber \\

739: \beta _i  &\equiv 

740:  \textstyle{\sum_{j\ne i} \left[(\bJ + \bB)_{ij} \omega \sin (k(j-i))

741:       + \tilde\bR_{ij} \cos (k(j-i))\right]}

742: \nonumber \end{align}

743: The term $\alpha _i\dot \bE_i$ when $\alpha_i>0$ feeds oscillation

744: energy into the {$i^{th}$ (receiving)} oscillator,

745: \wasred{causing {}{emergent} oscillations

746: in {}{ coupled damped} oscillators.}

747: We divide {}{$\alpha_i =\alpha_{i,\text{desc}}+\alpha_{i,\text{asc}}$}

748:   into the {}{descending and ascending}

749:   parts, with summations over  {}{$\sum_{j<i}$

750: and $\sum_{j>i}$} respectively.

751: Hence, for $i=1$, $\alpha_1 = \alpha_{i,\text{asc}}$; for  $i=N$,

752: $\alpha_N = \alpha_{i,\text{desc}}$, {and for $1\ll i \ll N$},

753: $\alpha_{i} = \alpha_{1} + \alpha_{N}$. Since the first

754: and last {}{segments}

755: oscillate due to the driving force from other

756: oscillators, $\alpha_{1}>0$ and $\alpha_{N}>0$. Consequently,

757: $\alpha_{1}< \alpha_{N/2}$ and $\alpha_{N}< \alpha_{N/2}$.

758: {}{Further}, since {}{descending connections

759:   are stronger},  \wasred{it is most

760: likely that, for $1\ll i \ll N$},  $\alpha_{i,\text{desc}}>

761: \alpha_{i,\text{asc}}$. Consequently, $\alpha _{1} < \alpha _{N}$.

762: {Hence, the rostral oscillator has a smaller amplitude than the caudal one,

763:   which in turn has a smaller amplitude than the central one

764:    {(Fig 2(E))}. 

765: Firing rate saturation and variations of the fixed point along the

766: body may obscure {}{this  pattern} in experimental data,

767: although body movements are indeed smallest near the head \cite{Paggett1998}.

768: \wasred{Similarly, oscillation amplitudes {}{will}

769:   be reduced in sections of 

770: spinal cords shorter than the typical lengths of inter-segment

771: connections, and

772: will eventually be zero in ever shorter sections, 

773: as observed in experiments \cite{Buchanan2001}.}

774: 

775: 

776: %

777: %

778: In summary,  {analysis of} a model  of the

779: CPG neural circuit in lampreys  {has given}

780: {}{new insights into} the neural connection structures {needed}

781: to

782: generate and control the swimming behaviour.

783: In particular, we predict that the contra-lateral connections between $\bC$

784: must be stronger than the self-excitatory connection strength of

785: the $\bE$ neurons; that the

786: $\bC$ neurons are more inhibited

787: (in their AC components) by $\bL$ neurons than excited by

788: the $\bE $ neurons;

789: and {have shown} how different swimming regimes can be selected by

790: scaling the strengths of the various neural connections without changing 

791:  \wasred{the} connection patterns.

792: Our framework should help to provide further

793: insights into CPGs of animal locomotion. 

794: %

795: 

796: {\bf Acknowledgements}: 

797: We thank Peter Dayan for  \wasred{comments,  Carl van Vreeswijk for

798:   {}{help with literature}, and

799: the Gatsby Charitable Foundation for support.}

800: %

801: %

802: \begin{thebibliography}{apssamp}

803: \small{

804: %

805: \bibitem{Grillner2002}Grillner S, Wallen P. 

806: %

807: Brain Res Rev.2002;40(1-3):92-106.

808: \bibitem{Buchanan2001} J.T.Buchanan 

809: %

810: Progr. of Neurobiology 63,

811:   2001,p.441-446. 

812: and J. Neurophysiol. 1999, 81(5), P2037-45.

813: %

814: \bibitem{McClellan1998}McClellan AD, Hagevik A.

815: %

816: %

817: %

818: Exp.Brain Res.1998,126:93-108

819: %

820: %

821: %

822: %

823: %

824: %

825: \bibitem{CohenEtal82}

826:  \wasred{Cohen AH, Holmes PJ, and Rand RH

827: %

828: %

829: J. Math. Biol. (1982) 13(3):345-69.}

830: \bibitem{E&K} Ermentrout G B, Kopell N. 

831: %

832: %

833: Comm. Pure

834: Appl. Math.

835:  (1986), 39(5), 623-660; Kopell N, Ermentrout G B. 

836: %

837: %

838:  SIAM J. Appl. Math.

839: (1994), 54(2), 478-507; Kopell N, Ermentrout G B. 

840: %

841: %

842: SIAM

843:   J. Appl. Math. 50(4),1014-10522.

844: \bibitem{Taylor1998} Taylor D, Holmes P, J. Math. Biol. (1998) 37, 419-446.

845: \bibitem{Jung} Jung R, Kiemel T, \& Cohen A H. J. Neurophysiol. (1996)

846:   75(3). 1074-86

847: %

848: %

849: \bibitem{Kozlov2002}

850: Kozlov AK, Ullen F, Fagerstedt P, Aurell E, Lansner A, Grillner S

851: %

852:  Biol Cybern 2002 86(1):1-14

853:  \wasred{\bibitem{McClellanHagevik1997}

854: McClellan AD, Hagevik A.  J Neurophysiol. 1997 78(1): 214-28.}

855: \bibitem{Miller1998} Miller W L, Sigvardt K A. 

856: %

857: %

858: J. Neurosci. Methods 1998, 80:113-128.

859: \bibitem{Paggett1998} Paggett K C, Gupta V, McClellan A D. Exp. Brain Res. 1998 119:213-223. 

860: }

861: 

862: \end{thebibliography}

863: %

864: 

865: 

866: 

867: \end{document}

868: 

869: 

870: 

871: 

872: