Pe has two main types of uses: a lexical one, as a spatial, viz. temporal, P which has also developed various other more abstract but still lexical meanings, and a grammatical one, as a DOM marker. Both are attested in the first texts of the 16th century, the second one appears to be in its first stages of grammaticalization. A spatial P evolving into a DOM marker is cross-linguistically a common phenomenon. Still, this particular grammaticalization raises several questions.
First, pe continues, through a metathesis (i.e., pre, the most frequent form in Old Romanian (OR)), Latin per (Ciorănescu 2005, a.o.). However, Latin per was initially used as a kind of perlative marker, i.e., it
expressed motion of the Trajector ACROSS or ABOUT the Landmark, whereas Romanian pre has the meaning ON a Landmark already in the earliest texts. While the originally spatial use of per in Latin extended to denoting temporal duration and its secondary uses include expression of INSTRUMENT (and also the AGENT in the passive), CAUSE and APPEAL, it never indicated location ON a Landmark (Brucale and Mocciaro 2011). This is also true of its reflexes in all other Romance languages.
Second, DOM markers in other Romance languages have developed similarly from a spatial P, a reflex of ¬ad, whose basic meaning of location NEXT TO developed into a direction marker with human landmarks (Luraghi 2011:215). Then, due to the low frequency of use of the morphological dative (Smith 2011:278-9), progressively grammaticalized into an analytical dative case marker well attested in Vulgar Latin but present already in the earliest texts (Väänänen 1981:113). The increasing number of verbs whose construction required a dative argument (often animate and human) is believed to have contributed to the reanalysis of the construction as an accusative marker used with referential human NPs. This ultimately converted it into a DOM marker in several Romance languages (Sornicola 2011: 36 ff). This pathway is straightforward since the grammaticalization of a P indicating direction into a dative marker is cross-linguistically well-attested, and subsequent extension of dative marking to semantically accusative (referential human) arguments is conceptually unproblematic. The second question to be answered then is why Romanian did not follow this pathway. What is particularly interesting is that p(r)e didn’t go through a similar dative stage.
Finally, pe grammaticalized into a DOM marker only in Daco-Romanian, whereas no authentic DOM (i.e., not influenced by schooling) has been reported for the other Balkan Romance (BR) varieties in close contact with Slavic languages (Friedman 2008, Hill 2013). This is all the more surprising since, in some of these Slavic varieties, P indicating location ON is used as a dative, genitive, and DOM marker (Adamou 2009).