The Vedanta Sutras, commentary by Sankaracharya (SBE34), tr. by George Thibaut  at sacred-texts.com
12. (The pañkaganâh are) the breath and so on, (as is seen) from the complementary passage.
The mantra in which the pañkaganâh are mentioned is followed by another one in which breath and four other things are mentioned for the purpose of describing the nature of Brahman. 'They who know the breath of breath, the eye of the eye, the ear of the ear, the food of food, the mind of mind 1.' Hence we conclude, on the ground of proximity, that the five-people are the beings mentioned in this latter mantra.--But how, the Sânkhya asks, can the word 'people' be applied to the breath, the eye, the ear, and so on?--How, we ask in return, can it be
applied to your categories? In both cases the common meaning of the word 'people' has to be disregarded; but in favour of our explanation is the fact that the breath, the eye, and so on, are mentioned in a complementary passage. The breath, the eye, &c. may be denoted by the word 'people' because they are connected with people. Moreover, we find the word 'person,' which means as much as 'people,' applied to the prânas in the passage, 'These are the five persons of Brahman' (Kh. Up. III, 13, 6); and another passage runs, 'Breath is father, breath is mother,' &c. (Kh. Up. VII, 15, 1). And, owing to the force of composition, there is no objection to the compound being taken in its settled conventional meaning 1.--But how can the conventional meaning be had recourse to, if there is no previous use of the word in that meaning?--That may be done, we reply, just as in the case of udbhid and similar words 2. We often infer that a word of unknown meaning refers to some known thing because it is used in connexion with the latter. So, for instance, in the case of the following words: 'He is to sacrifice with the udbhid; he cuts the yûpa; he makes the vedi.' Analogously we conclude that the term pañkaganâh, which, from the grammatical rule quoted, is known to be a name, and which therefore demands a thing of which it is the name, denotes the breath, the eye, and so on, which are connected with it through their being mentioned in a complementary passage.--Some commentators explain the word pañkaganâh
to mean the Gods, the Fathers, the Gandharvas, the Asuras, and the Rakshas. Others, again, think that the four castes together with the Nishâdas are meant. Again, some scriptural passage (Rig-veda Samh. VIII, 53, 7) speaks of the tribe of 'the five-people,' meaning thereby the created beings in general; and this latter explanation also might be applied to the passage under discussion. The teacher (the Sûtrakâra), on the other hand, aiming at showing that the passage does not refer to the twenty-five categories of the Sânkhyas, declares that on the ground of the complementary passage breath, &c. have to be understood.
Well, let it then be granted that the five-people mentioned in the Mâdhyandina-text are breath, &c. since that text mentions food also (and so makes up the number five). But how shall we interpret the Kânva-text which does not mention food (and thus altogether speaks of four things only)?--To this question the next Sûtra replies.
260:1 So in the Mâdhyandina recension of the Upanishad; the Kânva recension has not the clause 'the food of food.'
261:1 This in answer to the Sânkhya who objects to gana when applied to the prâna, &c. being interpreted with the help of lakshanâ; while if referred to the pradhâna, &c. it may be explained to have a direct meaning, on the ground of yaugika interpretation (the pradhâna being gana because it produces, the mahat &c. being gana because they are produced). The Vedântin points out that the compound pañkaganâh has its own rûdhi-meaning, just as asvakarna, literally horse-ear, which conventionally denotes a certain plant.
261:2 We infer that udbhid is the name of a sacrifice because it is mentioned in connexion with the act of sacrificing; we infer that the yûpa is a wooden post because it is said to be cut, and so on.