My own ideas on improving breadth and depth of role, however counter-intuitive it might first seem, was to actually put the effective tools of each role into the hands of anyone and everyone. That means that DPS have tools for mitigation on other's behalf and survival in their own rights. However, these would not be slotted skills; rather, they were given through universal undermechanics, tied into how damage and enmity would work.
Let's start with the idea of enmity. At present, enmity is generated at a fixed rate from throughput (damage or healing dealt), which can then be multiplied through a buff or particular skills (tank enmity combos), or divided at separate rates for healing and overhealing. But this lends nothing to (a) player control without a tank except by paring back one's damage or through the use of CDs that are useful at all times, but most effective during highest throughput-per-second windows (making them effectively just a tacked-on spare DPS CD), or (b) any tank beyond their choice of enmity or "other" skill, (c) the idea of organic tank-swapping.
What I'd suggest instead, therefore, is a revision to mob AI that allows for manipulation of enmity. Not only would AI be able to have non-standard modifiers, adjusting the way they react to throughputs over the course and conditions of a fight, but players could manipulate who receives that attention. Moreover, a mob needn't be limited to a single target or focus; a tanking player could potentially throw in additional threat at a well timed moment to distract a tail-swiping mob away from allies forced to move through its tail section when that swipe could be lethal, whether directly or indirectly. But moreover, enmity could be shifted between players through angular stacking, well-timed burst enmity during positional swaps, etc. I realize I'm not being terribly concrete, here, but while the patterns or total number of mechanics necessary make up only a short list (table, mode, conditions, focus), their implications and likely variance have filled up more than a page of my notes.
In addition to this, for damage to be raised without obligating a tank to a point that they feel like a gimmick requirement, there would need to be a way to distinguish between immediate and effectively cumulative damage intake. Ideally, I also wanted to tie this system into tools universally usable for external (suppressive) mitigation. To this extent, I would suggest a Stagger system, a percentile throughput system by which damage dealt reduces target outputs and/or increases intake via a continuously and dynamically fading penalty. Technically, this would be broken into applications of Force* and Pain* (*tentative terms), whereby Stagger is dealt via Force, which is a modifier off of damage, and fades at a rate determined by Pain, also a modifier determined by damage. This allows for synergy and coordination, increasing the need for focused and tactically spread pressure.
What this would mean on the player's end is that damage received would make it easier for yet more damage to be received, while also reducing the afflicted player's throughput produceable. This means that enemy attacks can be impactful without outright one- or two-shotting non-tanks, while tanks in turn see advantage simply due to the formulas for Stagger shared by both mobs and players, and because of their large HP pools and stacking ability for self-mitigation (rather than being limited typically to just suppression and avoidance).
The last couple things are an increased value on avoidance and its opposite, interception.
Consider what would happen if: (1) virtually all attacks were AoE, though still only the strongest were necessarily zone-marked, and (2) all AoEs faced cumulative mitigation — meaning that the damage absorbed by a first player reduces the damage done to the next, and then cumulatively to the third, and so on — or only struck the first in line, and (3) one (generally just tanks) could have a secondary, larger hitbox by which they intercept attacks that would otherwise hit any allied hitbox within.
For starters, a highly mobile player could avoid a greater portion of damage, but more importantly, even if one does not have threat on a mob or the mob literally cannot be focused onto just one player, one can still "tank" the given outgoing damage. Altogether, this makes tanking more about thwarting the enemy's offense — and as a group effort, at that — than simply stacking enmity values in order to meat-shield the attacks, and greatly opens up the variety of mob behavior scripts or AI possible without costing players vital control.
This doesn't yet begin to go into revisions to RNG mitigation (dodge, block, parry), new mechanisms like Guard or Cleave, or combo revisions, but this should at least summarize the relevant bits.