Since you asked about information, yes. The modulation appears on every harmonic, as sidebands mixed in above and below the harmonic.
Only the fundamental is useful, because it resembles AM. If the PWM is biased at 50% duty and modulated above and below there, the 2nd harmonic averages zero, and the sidebands around it are doubled (which means it's 100% distortion, which shouldn't be surprising since this is around a harmonic!). All harmonics go similarly, so the 3rd harmonic has tripled audio, and so on. At some point, the multiplied audio overlaps itself from adjacent harmonics, and the HF spectrum begins to resemble noise with harmonic spikes on it, rather than spikes with sidebands.
For example, a 1MHz carrier with 20kHz BW audio has:
0-20kHz: audio (baseband)
980kHz to 1020kHz: audio (sidebands above and below), 1000kHz carrier
1960kHz to 2040kHz: doubled audio (sidebands), 2000kHz null harmonic (if biased at 50%)
2940kHz to 3060kHz: tripled audio (sidebands), 3000kHz 1/3 amplitude harmonic (consistent with the average square wave form)
...
24MHz harmonic +/- 480kHz (24x audio) sidebands
25MHz harmonic +/- 500kHz (25x audio) sidebands
26MHz harmonic +/- 520kHz (26x audio) sidebands
Note that at this point, the sidebands overlap, so it starts to look more broadly noisy.
Note also that, although in theory, even harmonics are always null (because of the 50% bias condition), in practice, they will be present due to the imperfect matching of rising and falling edges, and imperfect 50% balance.
The harmonics themselves taper off as 1/N, until frequencies on the order of the rise/fall time, where they drop off as 1/N^2 or steeper. So a waveform with 10ns rise time will have notable harmonics out to 200MHz, then dropping off more rapidly beyond there.
Audio can still be recovered, even from higher harmonic sidebands, but a nonlinear recovery is required, because of the multiplier. It could be done with DSP, but isn't tractable with analog circuitry.
As for filters, note that HP * LP = LP * HP = BP (bandpass). It doesn't matter which order you put them in, or they can be mixed together in place (a typical BP filter prototype replaces the L's and C's of a LP design with L+C and L||C pairs, respectively). For an LC filter, matching is still required, so that you cannot simply wire them together directly, but some matching or termination impedance has to be introduced between them. For an active (op-amp) filter, matching doesn't matter.
For power signals, rather than informational purposes -- note that only baseband (LP recovered audio) and the fundamental (AM) matter. These will recover useful power with the respective filter type (LP or BP), while the remaining signal power is reflected back to the amplifier (hopefully without causing high pulsed currents, which is bad for amplifiers).
Tim