Using PNP instead of NPN merely requires one to make circuit upside down and reverse current direction, if this is true then why don't we just do everything with NPN transistors? Are there any applications where we can only use one of the two types? Why?
There have been times when transistors with specific characteristics were feasible in NPN, or PNP, but not vice-versa.
Two examples :
The 2N3055 power transistor was NPN, at a time when PNP transistors of the same power were simply not available (115W I think, later pushed to 150W).
That led to the "output triple" - a combination of 3 transistors that could replace either NPN or PNP output transistors, with the power handled in either case by the 2N3055. This was used in the famous Quad 303 amplifier of the early 1970s.
Another example : the BC214 PNP transistor could achieve lower noise in a microphone amplifier or audio input stage than contemporary NPN transistors.
With further developments in process technology, these considerations are less important than they were.
For many years, we did use only NPN transistors. Well actually, they weren't called transistors but "vacuum tubes". It would have often been convenient to have the complement of NPN or N-channel vacuum tubes, but those didn't exist. So yes, it's possible. You could do similar things today by designing a audio power amplifier, for example, using only NPN or only PNP transistors.
However, having complementary parts can be very useful and allows for circuit topologies not possible using just one polarity. Think of something like a audio power amplifier that needs to drive its output equally above and below ground. Individual transistors pull in one direction, and the rest of the circuit has to pull in the other direction if you want the signal to go that way. It shouldn't be hard to imagine that you can use one polarity transistor to pull high and the opposite polarity to pull low. The tops and bottom halves of such circuits are then mirror images of each other about ground, which requires flipping the polarity of the transistors. The output stages of most transistor audio power amplifiers exploit this NPN/PNP symmetry and really do work this way.
Having both pnp and npn transistors gives designers more flexability .Remember that all valves were N channel or npn .The filament dont emit positrons.Electrons are more mobile than holes so one would expect npn to give better performance .For Si BJTs the differences are very noticable at the high power area .Small signal GP transistors perform the same regardless of pnp or npn .Ge transistors are mainly pnp due to fabrication issues .In fact the npn types were worse .
If you go back to the early 1960's we only had Germanium transistors and they were all PNP; circuit diagrams were drawn with the -ve supply at the top. To provide enough power for a speaker the standard six transistor radio used transformer coupling to provide a push-pull output. Germanium was very prone to thermal runaway so transistor power amps didn't come into their own until the silicon 2N3055. Having PNP and NPN with matched characteristics makes symmetrical design so much easier.
One application where you can only use one type is when you implement the BJT as a "parasitic" PNP in a standard CMOS process. In a typical CMOS process NPNs are impossible to implement but a useful PNP can be made in which the collector is implemented in the substrate, the base is an n-well, and the collector is a p diffusion within that n-well. For more detail see pg. 3 of the following: notes on IC biasing This is extremely useful in the implementation of temperature-insensitive bias circuits, most notably the bandgap reference. Bandgap references. Note that a CMOS implementation using substrate PNPs must have the collector at ground, but the circuit in the link can be recast.