Table 1 Fit of demographic models to the observed data

Dataset	SbfI(1–3)		Dataset	SbfI(1–4)		Dataset	SbfI(1–3,6)		SbfI(1–3,7)
Model	ΔLL	ΔAIC	Model	ΔLL	ΔAIC	Model	ΔLL	ΔAIC	ΔLL	ΔAIC
3a*	79	−7*	3Lo	1343	−5313	3So	1145	−4435	978	−1963
3b	581	−2317	3Li	1513	−6095	3Si	1257	−4953	1012	−2121
3c	598	−2398	3Lv	600	−1892	3Sv	578	−1826	968	−1918
3d	503	−1959	3Lvx*	191	−14*	3Svx*	181	−4*	551	−2*
			3Lvo	222	−153	3Svo	197	−73	573	−105
			3Lvi*	188	0*	3Svi*	181	0*	550	0*

Dataset	SbfI(1–4,6)		SbfI(1–4,7)		NsiI(1–3,5,7)		NsiI(1–3,5,9)		NsiI(1–3,5,8)
Model	ΔLL	ΔAIC	ΔLL	ΔAIC	ΔLL	ΔAIC	ΔLL	ΔAIC	ΔLL	ΔAIC
PD	650	−1063	1532	−1304	184	−313	154	−249	411	−789
EVa	642	−1028	1535	−1321	185	−318	154	−253	414	−809
EVb	647	−1053	1528	−1289	182	−309	152	−244	424	−853
HOax*	437	−89	1421	−799	159	−205	104	−25*	306	−313
HOai*	440	−101	1420	−790	160	−205	99	0*	343	−483
HOao	461	−199	1428	−827	175	−278	124	−117	348	−503
HObx*	424	−29*	1337	−411	116	−7*	142	−199	435	−907
HObi*	418	0*	1337	−410	117	−9*	138	−178	440	−929
HObo*	453	−160	1355	−491	122	−31*	144	−208	435	−907
HOcx*	421	−20*	1249	−12*	113	0*	101	−18*	237	0*
HOci*	417	−1*	1247	0*	119	−20*	129	−141	279	−189
HOco*	454	−170	1283	−166	122	−37*	142	−203	246	−41*
SCi	711	−1346	1436	−860	509	−1811	236	−626	776	−2472
SCo	784	−1682	1529	−1288	434	−1465	237	−631	551	−1435

Best-fitting models and models with very similar likelihood are marked with an asterisk (*). Shown are log₁₀ likelihood differences between observed and expected site-frequency spectra (ΔLL) and difference in Akaike information criterion (ΔAIC) between the best and all models for a given dataset. For NsiI-data, the three West European lineages were modeled as unsampled (‘ghost’) populations, with parameters for the latter fixed to best estimates of model 3a (Supplementary Fig. 6). Numbers in brackets indicate populations used: West (Rhone: 1:FRS4, Rhine: 2:AGS1) and East European (3:PLS1), Lake Constance lake (4:L1, 5:^rROM) and stream populations (6:S2, 7:NID, 8:GRA, 9:BOH). Source data are provided as a Source Data file

Search