The gemara has a discussion about kinyan agav whether the objects have to be on the land. There is a general machlokes rishonim whether kinyan agav itself is d'oraysa or d'rabonon. Tosafos in Baba Kama 12a holds that it is only d'rabonon, therefore one who acquires an eved cannot be koneh metaltilin with him since for d'rabonon things an eved is not considered like property. However, Rashi in Gittin 77b indicates that agav would work to acquire a gett and be divorced m'doraysa, implying that it is d'oraysa. Similarly, Tosafos in kiddushin 5a d.h. sh'kain, suggest using agav to refute a d'oraysa kal v'chomer, implying that it is d'oraysa.
The gemara seems to understand that the kinyan that Rabban Gamliel used to be makneh the ma'aser was a kinyan agav (the only issue is whether we can prove from there that the metaltilin needs to be tziburin). Both Rashi and Tosafos seems to agree that the acquisition must be binding m'doraysa. Rashi assumes that the purpose was to be mafrish to avoid an issur of eating tevel which implies that it was binding m'doraysa. Tosafos assumes that the issue wasn't the separating, rather the distributing to fulfill the mitzvah of bi'ur, which also implies that it had to be binding m'doraysa. This seems to be a strong question on the opinion of Tosafos in Baba kamma who says that agav is only a kinyan d'rabonon.
The ketzos hachoshen (202:5) deals with a similar question from the gemara in baba metzia 46a that implies that kinyan agav works to avoid paying a chomesh, which implies that it is binding d'oryasa. He suggests that just as we find by ma'amad shelashton which is only a kinyan d'rabonon, that it works d'oryasa based on hefker beis din, the same thing could apply to kinyan agav. This would also answer the question from our gemara. He then suggests that even rashi who says that it works by gett may agree that it is only d'rabonon and works based on kol d'mekadesh ada'ata d'rabonon mekadesh.